{
 "retrieved": "2026-10-08",
 "tests": [
  {
   "id": "aa_intelligence_index",
   "name": "Artificial Analysis Intelligence Index v4.3.2",
   "creator": "Artificial Analysis",
   "year": 2026,
   "measures": "One synthetic score that averages ten independent tests of agents, coding, scientific reasoning and general knowledge.",
   "items": "10 evaluations (AA-Briefcase, GDPval-AA v2.1, AutomationBench-AA, Terminal-Bench 4.0, SciCode, AA-LCR, AA-Omniscience, HLE, GDP.pdf, CritPt)",
   "format": "Composite index (weighted average: Agents 30%, Coding 20%, Scientific reasoning 20%, General 30%)",
   "unit": "index points",
   "human_baseline": null,
   "saturation": "Not saturated: top model 57.6 (Claude Opus 5.5, max effort) on 2026-10-06, unchanged on 2026-10-08.",
   "known_issues": "Version changes often (v4.0 Jan 2026 to v4.3.2 Sep-Oct 2026); scores are not comparable across versions. Each reasoning-effort setting is scored separately. English, mostly text.",
   "url": "https://artificialanalysis.ai/methodology/intelligence-benchmarking"
  },
  {
   "id": "epoch_capabilities_index",
   "name": "Epoch Capabilities Index (ECI)",
   "creator": "Epoch AI",
   "year": null,
   "measures": "One capability score fitted across dozens of benchmarks, so that models tested on different benchmarks can be compared on one scale.",
   "items": null,
   "format": "Statistical composite (item-response-style fit over many benchmarks)",
   "unit": "ECI points (with 90% CI)",
   "human_baseline": null,
   "saturation": "Not saturated (open scale).",
   "known_issues": "Depends on which benchmarks each model has results for; very new models (e.g. Gemini 4 Argon, MiMo-V2.6-Pro, Mistral Large 4) not yet scored.",
   "url": "https://epoch.ai/benchmarks"
  },
  {
   "id": "lmarena_text",
   "name": "LMArena Text Arena (overall)",
   "creator": "LMArena",
   "year": null,
   "measures": "Which answer people prefer in blind side-by-side chats, turned into a rating.",
   "items": "8,626,731 votes across 413 models (snapshot of 2026-10-02)",
   "format": "Pairwise human preference votes",
   "unit": "Elo-style score",
   "human_baseline": null,
   "saturation": "Not saturated, but the top 20 models sit within about 40 points with overlapping confidence intervals.",
   "known_issues": "Measures preference, not correctness; sensitive to style and length (style control applied by default); new models have few votes and wide intervals (marked preliminary). The site now redirects from lmarena.ai to arena.ai ('formerly known as LMArena'); no new snapshot after 2026-10-02 as of 2026-10-08.",
   "url": "https://lmarena.ai/leaderboard/text"
  },
  {
   "id": "gpqa_diamond",
   "name": "GPQA Diamond",
   "creator": "David Rein, Samuel R. Bowman et al. (GPQA paper)",
   "year": 2023,
   "measures": "Very hard graduate-level multiple-choice questions in biology, physics and chemistry that cannot be answered by simple web search.",
   "items": "198",
   "format": "Multiple choice (4 options)",
   "unit": "% accuracy",
   "human_baseline": "PhD-level experts 65% (74% discounting clear mistakes); skilled non-experts with web access 34% (on the full GPQA set, per the paper abstract).",
   "saturation": "Near saturation: best scores around 95-96% (Epoch AI, Artificial Analysis). Removed from the AA Intelligence Index in v4.2.",
   "known_issues": "Small set (198 questions), so differences of 1-2 points are noise; the paper notes that experts reach 74% once clear mistakes they identified are discounted, so a few answers are disputable and the practical ceiling is below 100%.",
   "url": "https://arxiv.org/abs/2311.12022"
  },
  {
   "id": "hle",
   "name": "Humanity's Last Exam (HLE)",
   "creator": "Center for AI Safety and Scale AI",
   "year": 2025,
   "measures": "Expert-written closed-ended questions across dozens of academic subjects, designed to be the hardest academic exam for AI.",
   "items": "2,500 public questions (Artificial Analysis evaluates 2,158)",
   "format": "Exact-match short answers and multiple choice (24% MC, 14% need an image)",
   "unit": "% accuracy",
   "human_baseline": null,
   "saturation": "Not saturated: about 61% without tools (Claude Opus 5.5, AA); vendors report up to 67.7% with tools.",
   "known_issues": "With-tools and no-tools scores are mixed in announcements; answer grading uses an LLM judge (judge choice differs between AA, Scale and vendors); CAIS released an updated \"HLE-Rolling\" set on 2026-09-17.",
   "url": "https://lastexam.ai/"
  },
  {
   "id": "frontiermath_t1_3",
   "name": "FrontierMath Tiers 1-3 (v2, private)",
   "creator": "Epoch AI",
   "year": 2026,
   "measures": "Unpublished research-level mathematics problems with automatically checkable answers.",
   "items": null,
   "format": "Numeric / symbolic answers checked automatically",
   "unit": "% accuracy",
   "human_baseline": null,
   "saturation": "Approaching saturation for the very top (93.7% for GPT-6 Astra and GPT-6.1 Sol).",
   "known_issues": "Held privately and run only by Epoch AI, so results cannot be reproduced by third parties.",
   "url": "https://epoch.ai/benchmarks"
  },
  {
   "id": "frontiermath_t4",
   "name": "FrontierMath Tier 4 (v2, private)",
   "creator": "Epoch AI",
   "year": 2026,
   "measures": "The hardest tier of FrontierMath (research-level problems).",
   "items": null,
   "format": "Numeric / symbolic answers checked automatically",
   "unit": "% accuracy",
   "human_baseline": null,
   "saturation": "Saturating at the very top (GPT-6.1 Sol 100%, GPT-6 Astra 97.6%, Claude Opus 5.5 95%) but wide spread below.",
   "known_issues": "Very small number of problems, so each problem moves the score by several points.",
   "url": "https://epoch.ai/benchmarks"
  },
  {
   "id": "otis_mock_aime",
   "name": "OTIS Mock AIME 2024-2025",
   "creator": "Epoch AI (problems from the OTIS mock AIME)",
   "year": 2024,
   "measures": "Competition mathematics at the level of the American Invitational Mathematics Examination, on problems less likely to be in training data than the official AIME.",
   "items": null,
   "format": "Integer answers (0-999)",
   "unit": "% accuracy",
   "human_baseline": null,
   "saturation": "Saturated: several models at 100%.",
   "known_issues": "Used here as the AIME proxy because Artificial Analysis retired AIME 2025 and no independent AIME 2026 table covering current models was found.",
   "url": "https://epoch.ai/benchmarks"
  },
  {
   "id": "arc_agi_2",
   "name": "ARC-AGI-2",
   "creator": "ARC Prize Foundation",
   "year": 2025,
   "measures": "Visual grid puzzles that are easy for humans but require learning a new rule from a few examples.",
   "items": null,
   "format": "Grid output (pixel-exact), semi-private evaluation set",
   "unit": "% of tasks solved",
   "human_baseline": "Human panel 100% (ARC Prize data, about $17 per task).",
   "saturation": "Near saturation at the top (GPT-6 Astra 95%).",
   "known_issues": "Scores depend on reasoning effort and cost; ARC Prize reports cost per task alongside score.",
   "url": "https://arcprize.org/arc-agi/2"
  },
  {
   "id": "arc_agi_3",
   "name": "ARC-AGI-3",
   "creator": "ARC Prize Foundation",
   "year": 2026,
   "measures": "Interactive game environments where an agent must discover the goal and the rules by playing, scored on how efficiently it learns compared with humans.",
   "items": null,
   "format": "Interactive agentic games (no natural-language instructions)",
   "unit": "% score (100% = beats every game as efficiently as humans)",
   "human_baseline": "100% by construction (all environments are human-solvable).",
   "saturation": "Far from saturated with the standard harness (best 62.7%, GPT-6 Astra max effort).",
   "known_issues": "ARC Prize also publishes a \"Provider Adapter\" harness where some models score near 100%; the two harnesses must not be mixed.",
   "url": "https://arcprize.org/arc-agi/3"
  },
  {
   "id": "swe_bench_verified",
   "name": "SWE-bench Verified",
   "creator": "SWE-bench (swebench.com), human-verified subset",
   "year": 2024,
   "measures": "Fixing real GitHub issues from open-source repositories, checked by running unit tests against the generated patch.",
   "items": "500",
   "format": "Code patches (agentic)",
   "unit": "% issues resolved",
   "human_baseline": null,
   "saturation": "Saturated: 97.0% (Claude Opus 5, Vals.ai, mini-SWE-agent). Vals.ai no longer runs it on new releases and vendors stopped reporting it.",
   "known_issues": "Strong contamination risk (public repositories); scores vary a lot with the agent harness; newest models (Sept 2026) not tested.",
   "url": "https://www.swebench.com/"
  },
  {
   "id": "swe_bench_pro",
   "name": "SWE-Bench Pro (public set)",
   "creator": "Scale AI",
   "year": null,
   "measures": "Harder, contamination-resistant software engineering tasks from copyleft and private codebases.",
   "items": "1,865 tasks in total; public set 731, private set 276, held-out 858",
   "format": "Code patches (agentic)",
   "unit": "% resolved",
   "human_baseline": null,
   "saturation": "Not saturated, but the public SEAL leaderboard has not added models released after mid-2026 (latest entry Muse Spark 1.1).",
   "known_issues": "Most current frontier models have no SEAL result; vendors use their own harnesses.",
   "url": "https://scale.com/leaderboard/swe_bench_pro_public"
  },
  {
   "id": "terminal_bench_4",
   "name": "Terminal-Bench 4.0",
   "creator": "Stanford University researchers, Laude Institute and open-source community",
   "year": null,
   "measures": "Completing hard multi-step tasks in a command-line terminal (software, system administration, data, ML, security).",
   "items": "66",
   "format": "Agentic terminal tasks, each graded by its own test suite",
   "unit": "% tasks passed (pass@1)",
   "human_baseline": null,
   "saturation": "Not saturated (best 63.6% independent, 70.6% vendor-reported).",
   "known_issues": "Harness matters: AA uses mini-swe-agent (3 repeats), vendors use their own (Claude Code etc.); version 4.0 replaced 2.0/2.1 in 2026, so older Terminal-Bench numbers are not comparable.",
   "url": "https://github.com/harbor-framework/terminal-bench"
  },
  {
   "id": "tau3_banking",
   "name": "τ³-Banking (tau-bench family)",
   "creator": "Sierra",
   "year": 2026,
   "measures": "A customer-support agent that must search a large knowledge base and make correct account changes while talking with a simulated user.",
   "items": "97",
   "format": "Dual-control agent-user simulation with tools",
   "unit": "% pass@1 (5 repeats)",
   "human_baseline": null,
   "saturation": "Not saturated (best about 51%).",
   "known_issues": "Replaces τ²-Bench Telecom in AA reporting (τ²-Bench Telecom saturated above 95% for many models); removed from the AA Intelligence Index in v4.3 but still run as a standalone evaluation; not run for every new model.",
   "url": "https://github.com/sierra-research/tau2-bench"
  },
  {
   "id": "osworld_2",
   "name": "OSWorld 2.0 (binary completion)",
   "creator": "XLANG Lab (osworld-v2.xlang.ai)",
   "year": 2026,
   "measures": "Long, realistic computer-use workflows on a real desktop (about 1.6 hours median for a human).",
   "items": "108 workflows",
   "format": "Agentic computer use (GUI and tools), 500-step budget",
   "unit": "% tasks fully completed",
   "human_baseline": "Median human operation time about 1.6 hours per task (no human success rate published).",
   "saturation": "Far from saturated (best 31.4% binary on release v2026.08.08).",
   "known_issues": "Several releases (v2026.06.24, v2026.08.08, v2.1) with different scores; vendors report OSWorld 2.1 or subsets with partial scores; very few models on the public board.",
   "url": "https://osworld-v2.xlang.ai/"
  },
  {
   "id": "osworld_2_partial",
   "name": "OSWorld 2.0 / 2.1 (partial score)",
   "creator": "XLANG Lab (osworld-v2.xlang.ai)",
   "year": 2026,
   "measures": "Same tasks as OSWorld 2.0, with partial credit for sub-steps completed.",
   "items": "108 workflows",
   "format": "Agentic computer use",
   "unit": "% partial score",
   "human_baseline": null,
   "saturation": "Not saturated.",
   "known_issues": "Mixes independent (2.0 leaderboard) and vendor-reported (2.1, offline subset) values: compare only within the same version and source.",
   "url": "https://osworld-v2.xlang.ai/"
  },
  {
   "id": "metr_time_horizon_50",
   "name": "METR 50% task-completion time horizon",
   "creator": "METR",
   "year": 2025,
   "measures": "The length of software, ML and cybersecurity tasks (in human-expert time) that an AI agent completes with 50% probability.",
   "items": "Over a hundred tasks (RE-Bench, HCAST and shorter tasks), Time Horizon 1.1 suite",
   "format": "Agentic tasks; logistic fit of success vs. human task duration",
   "unit": "minutes",
   "human_baseline": "Task lengths are calibrated on skilled professionals (about 5 years of experience on average).",
   "saturation": "Suite near its limit: METR states measurements above 16 hours are unreliable.",
   "known_issues": "No measurement published for any model released after April 2026 (last update 2026-05-08, Claude Mythos Preview at about 1,045 min); very wide confidence intervals.",
   "url": "https://metr.org/time-horizons"
  },
  {
   "id": "mmlu_pro",
   "name": "MMLU-Pro",
   "creator": "TIGER-Lab",
   "year": 2024,
   "measures": "Broad academic knowledge and reasoning in 14 subjects, harder than the original MMLU.",
   "items": "Over 12,000 questions",
   "format": "Multiple choice (10 options)",
   "unit": "% accuracy",
   "human_baseline": null,
   "saturation": "Saturated: Vals.ai no longer runs it on new releases; top 92.4% (Claude Fable 5.1). Removed from the AA index in v4.0.",
   "known_issues": "No independent results for models released after early September 2026.",
   "url": "https://arxiv.org/abs/2406.01574"
  },
  {
   "id": "livecodebench",
   "name": "LiveCodeBench",
   "creator": "LiveCodeBench team (livecodebench.github.io)",
   "year": 2024,
   "measures": "Competitive programming problems collected continuously from LeetCode, AtCoder and Codeforces to limit contamination.",
   "items": "Over 1,000 problems in v6 (May 2023-2025)",
   "format": "Code generation checked by hidden tests",
   "unit": "% pass@1",
   "human_baseline": null,
   "saturation": "Saturated on easy and medium splits; Vals.ai no longer runs it on new releases; top 90.5%.",
   "known_issues": "Results depend on problem date window and version; removed from the AA index in v4.0.",
   "url": "https://livecodebench.github.io/"
  },
  {
   "id": "simpleqa_verified",
   "name": "SimpleQA Verified",
   "creator": null,
   "year": 2025,
   "measures": "Short factual questions that test whether a model knows facts without inventing answers.",
   "items": null,
   "format": "Short open answers graded for correctness",
   "unit": "% accuracy",
   "human_baseline": null,
   "saturation": "Not saturated (best 75.6%, GPT-6 Astra).",
   "known_issues": "Measures parametric knowledge without web search; models that abstain more score lower on accuracy but hallucinate less.",
   "url": "https://epoch.ai/benchmarks"
  },
  {
   "id": "aa_omniscience",
   "name": "AA-Omniscience",
   "creator": "Artificial Analysis",
   "year": null,
   "measures": "Knowledge across six domains, rewarding correct answers and penalising confident wrong answers (hallucinations).",
   "items": "6,000",
   "format": "Open answers",
   "unit": "index points (can be negative)",
   "human_baseline": null,
   "saturation": "Not saturated (best 46.4).",
   "known_issues": "A negative score means the model gives more wrong answers than right ones instead of abstaining; part of the AA Intelligence Index.",
   "url": "https://artificialanalysis.ai/methodology/intelligence-benchmarking"
  },
  {
   "id": "apex_agents",
   "name": "APEX-Agents",
   "creator": "Mercor",
   "year": 2026,
   "measures": "Professional-services tasks completed by an agent working with files and tools, graded by rubric.",
   "items": null,
   "format": "Agentic professional tasks, rubric-graded",
   "unit": "% pass@1",
   "human_baseline": null,
   "saturation": "Not saturated (best 82.2%, Gemini 4 Argon).",
   "known_issues": "Run by the benchmark owner; values here are compiled by Epoch AI; standard errors of about 4-5 points.",
   "url": "https://epoch.ai/benchmarks"
  }
 ],
 "models": [
  {
   "id": "gpt-6-astra",
   "name": "GPT-6 Astra",
   "provider": "OpenAI",
   "release_date": "2026-09-03",
   "open_weights": false,
   "note": null
  },
  {
   "id": "gpt-6-1-sol",
   "name": "GPT-6.1 Sol",
   "provider": "OpenAI",
   "release_date": "2026-09-29",
   "open_weights": false,
   "note": null
  },
  {
   "id": "gpt-6-luna",
   "name": "GPT-6 Luna",
   "provider": "OpenAI",
   "release_date": "2026-09-22",
   "open_weights": false,
   "note": null
  },
  {
   "id": "gpt-5-6-terra",
   "name": "GPT-5.6 Terra",
   "provider": "OpenAI",
   "release_date": "2026-07-09",
   "open_weights": false,
   "note": null
  },
  {
   "id": "gpt-oss-120b",
   "name": "gpt-oss-120b",
   "provider": "OpenAI",
   "release_date": "2025-08-05",
   "open_weights": true,
   "note": null
  },
  {
   "id": "claude-opus-5-5",
   "name": "Claude Opus 5.5",
   "provider": "Anthropic",
   "release_date": "2026-09-22",
   "open_weights": false,
   "note": null
  },
  {
   "id": "claude-sonnet-5-5",
   "name": "Claude Sonnet 5.5",
   "provider": "Anthropic",
   "release_date": "2026-09-28",
   "open_weights": false,
   "note": null
  },
  {
   "id": "claude-fable-5-1",
   "name": "Claude Fable 5.1",
   "provider": "Anthropic",
   "release_date": "2026-09-01",
   "open_weights": false,
   "note": null
  },
  {
   "id": "claude-opus-5",
   "name": "Claude Opus 5",
   "provider": "Anthropic",
   "release_date": "2026-07-24",
   "open_weights": false,
   "note": "Marked deprecated/superseded on Artificial Analysis (replaced by Claude Opus 5.5)."
  },
  {
   "id": "claude-haiku-5-5",
   "name": "Claude Haiku 5.5",
   "provider": "Anthropic",
   "release_date": "2026-10-07",
   "open_weights": false,
   "note": "Released 2026-10-07 (Claude Platform release notes). Only Artificial Analysis had results on 2026-10-08; not yet on LMArena (snapshot 2026-10-02), Epoch AI, ARC Prize, Vals.ai or OSWorld 2.0."
  },
  {
   "id": "claude-haiku-4-5",
   "name": "Claude Haiku 4.5",
   "provider": "Anthropic",
   "release_date": "2025-10-15",
   "open_weights": false,
   "note": "Release date 2025-10-15 per Artificial Analysis and Epoch AI (API id claude-haiku-4-5-20251001). Labelled 'Legacy' by Anthropic since Claude Haiku 5.5 (2026-10-07)."
  },
  {
   "id": "gemini-4-argon",
   "name": "Gemini 4 Argon",
   "provider": "Google",
   "release_date": "2026-09-30",
   "open_weights": false,
   "note": "Very recent (2026-09-30); LMArena score is marked preliminary."
  },
  {
   "id": "gemini-3-8-flash",
   "name": "Gemini 3.8 Flash",
   "provider": "Google",
   "release_date": "2026-09-02",
   "open_weights": false,
   "note": null
  },
  {
   "id": "gemini-3-1-pro",
   "name": "Gemini 3.1 Pro (Preview)",
   "provider": "Google",
   "release_date": "2026-02-19",
   "open_weights": false,
   "note": "Still a preview model; superseded in Google line-up by Gemini 4 Argon and Gemini 3.8 Flash."
  },
  {
   "id": "gemini-3-5-flash-lite",
   "name": "Gemini 3.5 Flash-Lite",
   "provider": "Google",
   "release_date": "2026-07-21",
   "open_weights": false,
   "note": null
  },
  {
   "id": "gemma-4-31b",
   "name": "Gemma 4 31B",
   "provider": "Google",
   "release_date": "2026-04-02",
   "open_weights": true,
   "note": null
  },
  {
   "id": "muse-spark-1-3",
   "name": "Muse Spark 1.3",
   "provider": "Meta",
   "release_date": "2026-09-02",
   "open_weights": false,
   "note": null
  },
  {
   "id": "grok-4-7",
   "name": "Grok 4.7",
   "provider": "xAI (shown as SpaceXAI on leaderboards)",
   "release_date": "2026-09-21",
   "open_weights": false,
   "note": "Artificial Analysis and LMArena list the creator as \"SpaceXAI\"; Epoch AI and ARC Prize list xAI."
  },
  {
   "id": "mistral-large-4",
   "name": "Mistral Large 4 (Preview)",
   "provider": "Mistral AI",
   "release_date": "2026-10-06",
   "open_weights": false,
   "note": "Preview released on 2026-10-06 per Artificial Analysis; very few third-party results yet."
  },
  {
   "id": "mistral-medium-3-5",
   "name": "Mistral Medium 3.5",
   "provider": "Mistral AI",
   "release_date": "2026-04-29",
   "open_weights": true,
   "note": "Release date per Artificial Analysis (Epoch AI: 2026-04-28)."
  },
  {
   "id": "deepseek-v4-1-flash",
   "name": "DeepSeek V4.1 Flash",
   "provider": "DeepSeek",
   "release_date": "2026-09-10",
   "open_weights": true,
   "note": null
  },
  {
   "id": "deepseek-v4-pro-0813",
   "name": "DeepSeek V4 Pro (0813)",
   "provider": "DeepSeek",
   "release_date": "2026-08-13",
   "open_weights": true,
   "note": null
  },
  {
   "id": "qwen3-8-max",
   "name": "Qwen3.8 Max (0902)",
   "provider": "Alibaba (Qwen)",
   "release_date": "2026-09-02",
   "open_weights": false,
   "note": "Refresh of Qwen 3.8 Max (first released 2026-08-02 per Epoch AI). Vals.ai rows labelled \"Qwen 3.8 Max\" may refer to the earlier version."
  },
  {
   "id": "qwen3-8-27b",
   "name": "Qwen3.8 27B",
   "provider": "Alibaba (Qwen)",
   "release_date": "2026-08-14",
   "open_weights": true,
   "note": null
  },
  {
   "id": "kimi-k3",
   "name": "Kimi K3",
   "provider": "Moonshot AI",
   "release_date": "2026-07-16",
   "open_weights": true,
   "note": null
  },
  {
   "id": "glm-5-3",
   "name": "GLM-5.3",
   "provider": "Zhipu AI (Z.ai)",
   "release_date": "2026-08-18",
   "open_weights": true,
   "note": "Release date per Artificial Analysis (2026-08-18); Epoch AI lists 2026-08-14 and classes it closed weights, while AA and LMArena (MIT licence) list open weights."
  },
  {
   "id": "minimax-m3",
   "name": "MiniMax-M3",
   "provider": "MiniMax",
   "release_date": "2026-06-01",
   "open_weights": true,
   "note": null
  },
  {
   "id": "mimo-v2-6-pro",
   "name": "MiMo-V2.6-Pro",
   "provider": "Xiaomi",
   "release_date": "2026-09-21",
   "open_weights": true,
   "note": "Release date per Artificial Analysis."
  }
 ],
 "scores": [
  {
   "model": "gpt-6-astra",
   "test": "aa_intelligence_index",
   "value": 52.67,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6 Astra (Max)."
  },
  {
   "model": "gpt-6-astra",
   "test": "hle",
   "value": 54.68,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6 Astra (Max), without tools. OpenAI's launch page publishes 57.2% with tools, maximum over reasoning efforts (https://openai.com/index/gpt-6-astra/, read 2026-10-06). Epoch AI's compilation of the HLE leaderboard lists 54.8%."
  },
  {
   "model": "gpt-6-astra",
   "test": "terminal_bench_4",
   "value": 59.09,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6 Astra (Max). Vendor (OpenAI, as quoted by Anthropic): 57.9% at high effort; Google quotes 58.2%."
  },
  {
   "model": "gpt-6-astra",
   "test": "tau3_banking",
   "value": 41.44,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6 Astra (Max)."
  },
  {
   "model": "gpt-6-astra",
   "test": "aa_omniscience",
   "value": 43.4,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6 Astra (Max)."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "aa_intelligence_index",
   "value": 51.83,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6.1 Sol (Max)."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "hle",
   "value": 52.92,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6.1 Sol (Max)."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "terminal_bench_4",
   "value": 56.06,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6.1 Sol (Max)."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "aa_omniscience",
   "value": 41.53,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6.1 Sol (Max)."
  },
  {
   "model": "gpt-6-luna",
   "test": "aa_intelligence_index",
   "value": 38.12,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6 Luna (Max)."
  },
  {
   "model": "gpt-6-luna",
   "test": "hle",
   "value": 38.51,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6 Luna (Max)."
  },
  {
   "model": "gpt-6-luna",
   "test": "terminal_bench_4",
   "value": 12.63,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6 Luna (Max)."
  },
  {
   "model": "gpt-6-luna",
   "test": "aa_omniscience",
   "value": 0.65,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-6 Luna (Max)."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "aa_intelligence_index",
   "value": 42.08,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-5.6 Terra (Max)."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "hle",
   "value": 42.91,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-5.6 Terra (Max)."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "terminal_bench_4",
   "value": 35.35,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-5.6 Terra (Max)."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "tau3_banking",
   "value": 40.21,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-5.6 Terra (Max)."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "aa_omniscience",
   "value": 0.05,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GPT-5.6 Terra (Max)."
  },
  {
   "model": "gpt-oss-120b",
   "test": "aa_intelligence_index",
   "value": 11.6,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: gpt-oss-120b (High)."
  },
  {
   "model": "gpt-oss-120b",
   "test": "hle",
   "value": 19.6,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: gpt-oss-120b (High)."
  },
  {
   "model": "gpt-oss-120b",
   "test": "terminal_bench_4",
   "value": 0,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: gpt-oss-120b (High)."
  },
  {
   "model": "gpt-oss-120b",
   "test": "tau3_banking",
   "value": 12.78,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: gpt-oss-120b (High)."
  },
  {
   "model": "gpt-oss-120b",
   "test": "aa_omniscience",
   "value": -49.25,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: gpt-oss-120b (High)."
  },
  {
   "model": "claude-opus-5-5",
   "test": "aa_intelligence_index",
   "value": 57.62,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Opus 5.5 (Max, Default Fallback)."
  },
  {
   "model": "claude-opus-5-5",
   "test": "hle",
   "value": 61.35,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Opus 5.5 (Max, Default Fallback). Vendor (Anthropic): 67.7% with tools."
  },
  {
   "model": "claude-opus-5-5",
   "test": "terminal_bench_4",
   "value": 59.6,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Opus 5.5 (Max, Default Fallback). Vendor (Anthropic): 66.4% at xhigh effort, own harness."
  },
  {
   "model": "claude-opus-5-5",
   "test": "aa_omniscience",
   "value": 46.42,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Opus 5.5 (Max, Default Fallback)."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "aa_intelligence_index",
   "value": 56.0,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Sonnet 5.5 (Max, Default Fallback)."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "hle",
   "value": 54.96,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Sonnet 5.5 (Max, Default Fallback). Vendor (Anthropic): 64.5% with tools."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "terminal_bench_4",
   "value": 63.64,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Sonnet 5.5 (Max, Default Fallback). Vendor (Anthropic): 70.6%."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "aa_omniscience",
   "value": 32.3,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Sonnet 5.5 (Max, Default Fallback)."
  },
  {
   "model": "claude-fable-5-1",
   "test": "aa_intelligence_index",
   "value": 53.35,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Fable 5.1 (Max, Default Fallback)."
  },
  {
   "model": "claude-fable-5-1",
   "test": "hle",
   "value": 59.13,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Fable 5.1 (Max, Default Fallback). Vendor (Anthropic): 60.9% without tools, 65.0% with tools."
  },
  {
   "model": "claude-fable-5-1",
   "test": "terminal_bench_4",
   "value": 52.02,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Fable 5.1 (Max, Default Fallback). Vendor (Anthropic): 55.8%."
  },
  {
   "model": "claude-fable-5-1",
   "test": "tau3_banking",
   "value": 47.22,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Fable 5.1 (Max, Default Fallback)."
  },
  {
   "model": "claude-fable-5-1",
   "test": "aa_omniscience",
   "value": 43.45,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Fable 5.1 (Max, Default Fallback)."
  },
  {
   "model": "claude-opus-5",
   "test": "aa_intelligence_index",
   "value": 50.78,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Opus 5 (Max)."
  },
  {
   "model": "claude-opus-5",
   "test": "hle",
   "value": 54.87,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Opus 5 (Max). Vendor (Anthropic): 56.6% without tools, 63.6% with tools."
  },
  {
   "model": "claude-opus-5",
   "test": "terminal_bench_4",
   "value": 48.99,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Opus 5 (Max). Vendor (Anthropic): 52.3%; public tbench.ai leaderboard 51.8% (Claude Code harness), per Anthropic."
  },
  {
   "model": "claude-opus-5",
   "test": "tau3_banking",
   "value": 42.06,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Opus 5 (Max)."
  },
  {
   "model": "claude-opus-5",
   "test": "aa_omniscience",
   "value": 37.07,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Opus 5 (Max)."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "aa_intelligence_index",
   "value": 16.88,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude 4.5 Haiku (Reasoning)."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "hle",
   "value": 10.38,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude 4.5 Haiku (Reasoning)."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "terminal_bench_4",
   "value": 0,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude 4.5 Haiku (Reasoning)."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "tau3_banking",
   "value": 9.28,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude 4.5 Haiku (Reasoning)."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "aa_omniscience",
   "value": -4.37,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude 4.5 Haiku (Reasoning)."
  },
  {
   "model": "gemini-4-argon",
   "test": "aa_intelligence_index",
   "value": 52.56,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 4 Argon (High)."
  },
  {
   "model": "gemini-4-argon",
   "test": "hle",
   "value": 57.09,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 4 Argon (High)."
  },
  {
   "model": "gemini-4-argon",
   "test": "terminal_bench_4",
   "value": 57.07,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 4 Argon (High). Vendor (Google): 57.4%."
  },
  {
   "model": "gemini-4-argon",
   "test": "aa_omniscience",
   "value": 42.35,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 4 Argon (High)."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "aa_intelligence_index",
   "value": 40.93,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.8 Flash (High)."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "hle",
   "value": 47.82,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.8 Flash (High). Epoch AI's compilation of the HLE leaderboard lists 44.5%."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "terminal_bench_4",
   "value": 19.7,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.8 Flash (High)."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "tau3_banking",
   "value": 44.95,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.8 Flash (High)."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "aa_omniscience",
   "value": 29.55,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.8 Flash (High)."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "aa_intelligence_index",
   "value": 29.72,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.1 Pro Preview."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "hle",
   "value": 47.03,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.1 Pro Preview. Epoch AI's compilation of the HLE leaderboard lists 46.4% (thinking high)."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "terminal_bench_4",
   "value": 4.04,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.1 Pro Preview."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "tau3_banking",
   "value": 21.44,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.1 Pro Preview."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "aa_omniscience",
   "value": 31.88,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.1 Pro Preview."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "aa_intelligence_index",
   "value": 22.17,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.5 Flash-Lite."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "hle",
   "value": 18.81,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.5 Flash-Lite."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "terminal_bench_4",
   "value": 1.01,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.5 Flash-Lite."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "tau3_banking",
   "value": 17.53,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.5 Flash-Lite."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "aa_omniscience",
   "value": 5.23,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemini 3.5 Flash-Lite."
  },
  {
   "model": "gemma-4-31b",
   "test": "aa_intelligence_index",
   "value": 14.67,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemma 4 31B (Reasoning)."
  },
  {
   "model": "gemma-4-31b",
   "test": "hle",
   "value": 23.63,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemma 4 31B (Reasoning)."
  },
  {
   "model": "gemma-4-31b",
   "test": "terminal_bench_4",
   "value": 0,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemma 4 31B (Reasoning)."
  },
  {
   "model": "gemma-4-31b",
   "test": "tau3_banking",
   "value": 14.85,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemma 4 31B (Reasoning)."
  },
  {
   "model": "gemma-4-31b",
   "test": "aa_omniscience",
   "value": -47.93,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Gemma 4 31B (Reasoning)."
  },
  {
   "model": "muse-spark-1-3",
   "test": "aa_intelligence_index",
   "value": 48.09,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Muse Spark 1.3 (Max)."
  },
  {
   "model": "muse-spark-1-3",
   "test": "hle",
   "value": 48.7,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Muse Spark 1.3 (Max)."
  },
  {
   "model": "muse-spark-1-3",
   "test": "terminal_bench_4",
   "value": 33.33,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Muse Spark 1.3 (Max)."
  },
  {
   "model": "muse-spark-1-3",
   "test": "tau3_banking",
   "value": 50.52,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Muse Spark 1.3 (Max)."
  },
  {
   "model": "muse-spark-1-3",
   "test": "aa_omniscience",
   "value": 25,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Muse Spark 1.3 (Max)."
  },
  {
   "model": "grok-4-7",
   "test": "aa_intelligence_index",
   "value": 46.45,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Grok 4.7 (Xhigh)."
  },
  {
   "model": "grok-4-7",
   "test": "hle",
   "value": 43.14,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Grok 4.7 (Xhigh)."
  },
  {
   "model": "grok-4-7",
   "test": "terminal_bench_4",
   "value": 25.76,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Grok 4.7 (Xhigh)."
  },
  {
   "model": "grok-4-7",
   "test": "aa_omniscience",
   "value": 32.03,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Grok 4.7 (Xhigh)."
  },
  {
   "model": "mistral-large-4",
   "test": "aa_intelligence_index",
   "value": 38.38,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Large 4 Preview."
  },
  {
   "model": "mistral-large-4",
   "test": "hle",
   "value": 35.03,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Large 4 Preview."
  },
  {
   "model": "mistral-large-4",
   "test": "terminal_bench_4",
   "value": 26.77,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Large 4 Preview."
  },
  {
   "model": "mistral-large-4",
   "test": "aa_omniscience",
   "value": -5.3,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Large 4 Preview."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "aa_intelligence_index",
   "value": 14.19,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Medium 3.5."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "hle",
   "value": 13.76,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Medium 3.5."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "terminal_bench_4",
   "value": 0,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Medium 3.5."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "tau3_banking",
   "value": 15.05,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Medium 3.5."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "aa_omniscience",
   "value": -36.8,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Medium 3.5."
  },
  {
   "model": "deepseek-v4-1-flash",
   "test": "aa_intelligence_index",
   "value": 39.46,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: DeepSeek V4.1 Flash (Max)."
  },
  {
   "model": "deepseek-v4-1-flash",
   "test": "hle",
   "value": 39.25,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: DeepSeek V4.1 Flash (Max)."
  },
  {
   "model": "deepseek-v4-1-flash",
   "test": "terminal_bench_4",
   "value": 26.77,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: DeepSeek V4.1 Flash (Max)."
  },
  {
   "model": "deepseek-v4-1-flash",
   "test": "aa_omniscience",
   "value": -5.3,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: DeepSeek V4.1 Flash (Max)."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "aa_intelligence_index",
   "value": 36.0,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: DeepSeek V4 Pro 0813 (Max)."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "hle",
   "value": 41.01,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: DeepSeek V4 Pro 0813 (Max)."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "terminal_bench_4",
   "value": 14.14,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: DeepSeek V4 Pro 0813 (Max)."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "tau3_banking",
   "value": 39.59,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: DeepSeek V4 Pro 0813 (Max)."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "aa_omniscience",
   "value": 0.83,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: DeepSeek V4 Pro 0813 (Max)."
  },
  {
   "model": "qwen3-8-max",
   "test": "aa_intelligence_index",
   "value": 45.42,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 Max (0902)."
  },
  {
   "model": "qwen3-8-max",
   "test": "hle",
   "value": 43.1,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 Max (0902)."
  },
  {
   "model": "qwen3-8-max",
   "test": "terminal_bench_4",
   "value": 38.89,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 Max (0902)."
  },
  {
   "model": "qwen3-8-max",
   "test": "tau3_banking",
   "value": 47.84,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 Max (0902)."
  },
  {
   "model": "qwen3-8-max",
   "test": "aa_omniscience",
   "value": 11.98,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 Max (0902)."
  },
  {
   "model": "qwen3-8-27b",
   "test": "aa_intelligence_index",
   "value": 33.7,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 27B (Xhigh)."
  },
  {
   "model": "qwen3-8-27b",
   "test": "hle",
   "value": 33.92,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 27B (Xhigh)."
  },
  {
   "model": "qwen3-8-27b",
   "test": "terminal_bench_4",
   "value": 5.56,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 27B (Xhigh)."
  },
  {
   "model": "qwen3-8-27b",
   "test": "tau3_banking",
   "value": 48.04,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 27B (Xhigh)."
  },
  {
   "model": "qwen3-8-27b",
   "test": "aa_omniscience",
   "value": -9.98,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 27B (Xhigh)."
  },
  {
   "model": "kimi-k3",
   "test": "aa_intelligence_index",
   "value": 43.59,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Kimi K3 (Max)."
  },
  {
   "model": "kimi-k3",
   "test": "hle",
   "value": 46.9,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Kimi K3 (Max)."
  },
  {
   "model": "kimi-k3",
   "test": "terminal_bench_4",
   "value": 12.63,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Kimi K3 (Max)."
  },
  {
   "model": "kimi-k3",
   "test": "tau3_banking",
   "value": 45.98,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Kimi K3 (Max)."
  },
  {
   "model": "kimi-k3",
   "test": "aa_omniscience",
   "value": 19.7,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Kimi K3 (Max)."
  },
  {
   "model": "glm-5-3",
   "test": "aa_intelligence_index",
   "value": 44.78,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GLM-5.3 (Max)."
  },
  {
   "model": "glm-5-3",
   "test": "hle",
   "value": 42.26,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GLM-5.3 (Max)."
  },
  {
   "model": "glm-5-3",
   "test": "terminal_bench_4",
   "value": 41.92,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GLM-5.3 (Max)."
  },
  {
   "model": "glm-5-3",
   "test": "tau3_banking",
   "value": 50.31,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GLM-5.3 (Max)."
  },
  {
   "model": "glm-5-3",
   "test": "aa_omniscience",
   "value": 14.3,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: GLM-5.3 (Max)."
  },
  {
   "model": "minimax-m3",
   "test": "aa_intelligence_index",
   "value": 29.22,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: MiniMax-M3."
  },
  {
   "model": "minimax-m3",
   "test": "hle",
   "value": 38.97,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: MiniMax-M3."
  },
  {
   "model": "minimax-m3",
   "test": "terminal_bench_4",
   "value": 2.02,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: MiniMax-M3."
  },
  {
   "model": "minimax-m3",
   "test": "tau3_banking",
   "value": 15.26,
   "unit": "% pass@1",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: MiniMax-M3."
  },
  {
   "model": "minimax-m3",
   "test": "aa_omniscience",
   "value": 1.35,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: MiniMax-M3."
  },
  {
   "model": "mimo-v2-6-pro",
   "test": "aa_intelligence_index",
   "value": 46.32,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: MiMo-V2.6-Pro."
  },
  {
   "model": "mimo-v2-6-pro",
   "test": "hle",
   "value": 49.35,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: MiMo-V2.6-Pro."
  },
  {
   "model": "mimo-v2-6-pro",
   "test": "terminal_bench_4",
   "value": 34.85,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: MiMo-V2.6-Pro."
  },
  {
   "model": "mimo-v2-6-pro",
   "test": "aa_omniscience",
   "value": 8.38,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: MiMo-V2.6-Pro."
  },
  {
   "model": "gpt-6-astra",
   "test": "lmarena_text",
   "value": 1477,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gpt-6-astra-max\", rank 29 of 413, ±7 (95% CI), 9,156 votes. Default view (style control on)."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "lmarena_text",
   "value": 1483,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gpt-6.1-sol-max\", rank 21 of 413, ±11 (95% CI), 3,071 votes. Default view (style control on)."
  },
  {
   "model": "gpt-6-luna",
   "test": "lmarena_text",
   "value": 1443,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gpt-6-luna-max\", rank 90 of 413, ±7 (95% CI), 10,014 votes. Default view (style control on)."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "lmarena_text",
   "value": 1465,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gpt-5.6-terra-xhigh\", rank 51 of 413, ±4 (95% CI), 37,663 votes. Default view (style control on)."
  },
  {
   "model": "gpt-oss-120b",
   "test": "lmarena_text",
   "value": 1352,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gpt-oss-120b\", rank 212 of 413, ±4 (95% CI), 30,018 votes. Default view (style control on)."
  },
  {
   "model": "claude-opus-5-5",
   "test": "lmarena_text",
   "value": 1504,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"claude-opus-5.5-high\", rank 4 of 413, ±9 (95% CI), 4,552 votes. Default view (style control on)."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "lmarena_text",
   "value": 1471,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"claude-sonnet-5.5-xhigh\", rank 45 of 413, ±10 (95% CI), 3,145 votes. Default view (style control on)."
  },
  {
   "model": "claude-fable-5-1",
   "test": "lmarena_text",
   "value": 1501,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"claude-fable-5.1-max\", rank 6 of 413, ±6 (95% CI), 11,800 votes. Default view (style control on)."
  },
  {
   "model": "claude-opus-5",
   "test": "lmarena_text",
   "value": 1490,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"claude-opus-5-high\", rank 13 of 413, ±4 (95% CI), 59,482 votes. Default view (style control on)."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "lmarena_text",
   "value": 1414,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"claude-haiku-4-5-20251001\", rank 142 of 413, ±2 (95% CI), 143,402 votes. Default view (style control on)."
  },
  {
   "model": "gemini-4-argon",
   "test": "lmarena_text",
   "value": 1525,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gemini-4-argon-high\", rank 1 of 413, ±9 (95% CI), 4,932 votes, marked preliminary. Default view (style control on)."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "lmarena_text",
   "value": 1495,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gemini-3.8-flash-high\", rank 8 of 413, ±5 (95% CI), 26,298 votes, marked preliminary. Default view (style control on)."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "lmarena_text",
   "value": 1487,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gemini-3.1-pro-preview\", rank 18 of 413, ±3 (95% CI), 121,806 votes. Default view (style control on)."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "lmarena_text",
   "value": 1455,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gemini-3.5-flash-lite\", rank 71 of 413, ±4 (95% CI), 35,763 votes. Default view (style control on)."
  },
  {
   "model": "gemma-4-31b",
   "test": "lmarena_text",
   "value": 1453,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"gemma-4-31b\", rank 76 of 413, ±7 (95% CI), 6,133 votes. Default view (style control on)."
  },
  {
   "model": "muse-spark-1-3",
   "test": "lmarena_text",
   "value": 1494,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"muse-spark-1.3-max\", rank 9 of 413, ±6 (95% CI), 12,343 votes. Default view (style control on)."
  },
  {
   "model": "grok-4-7",
   "test": "lmarena_text",
   "value": 1442,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"grok-4.7-xhigh\", rank 91 of 413, ±8 (95% CI), 6,405 votes. Default view (style control on)."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "lmarena_text",
   "value": 1427,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"mistral-medium-3.5\", rank 115 of 413, ±6 (95% CI), 11,757 votes. Default view (style control on)."
  },
  {
   "model": "deepseek-v4-1-flash",
   "test": "lmarena_text",
   "value": 1474,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"deepseek-v4.1-flash-max\", rank 38 of 413, ±7 (95% CI), 8,728 votes. Default view (style control on)."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "lmarena_text",
   "value": 1464,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"deepseek-v4-pro-high-20260813\", rank 56 of 413, ±7 (95% CI), 9,780 votes. Default view (style control on)."
  },
  {
   "model": "qwen3-8-max",
   "test": "lmarena_text",
   "value": 1482,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"qwen3.8-max\", rank 23 of 413, ±5 (95% CI), 23,353 votes. Default view (style control on)."
  },
  {
   "model": "qwen3-8-27b",
   "test": "lmarena_text",
   "value": 1438,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"qwen3.8-27b\", rank 99 of 413, ±5 (95% CI), 18,462 votes. Default view (style control on)."
  },
  {
   "model": "kimi-k3",
   "test": "lmarena_text",
   "value": 1488,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"kimi-k3-max\", rank 16 of 413, ±5 (95% CI), 28,280 votes. Default view (style control on)."
  },
  {
   "model": "glm-5-3",
   "test": "lmarena_text",
   "value": 1478,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"glm-5.3-max\", rank 27 of 413, ±6 (95% CI), 17,857 votes. Default view (style control on)."
  },
  {
   "model": "minimax-m3",
   "test": "lmarena_text",
   "value": 1440,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"minimax-m3\", rank 96 of 413, ±4 (95% CI), 58,757 votes. Default view (style control on)."
  },
  {
   "model": "mimo-v2-6-pro",
   "test": "lmarena_text",
   "value": 1480,
   "unit": "Elo-style score",
   "source": "LMArena Text Arena",
   "url": "https://lmarena.ai/leaderboard/text",
   "date": "2026-10-02",
   "kind": "independent",
   "note": "Entry \"mimo-v2.6-pro\", rank 26 of 413, ±9 (95% CI), 4,056 votes. Default view (style control on)."
  },
  {
   "model": "gpt-6-astra",
   "test": "epoch_capabilities_index",
   "value": 166.45,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 163.34-171.46. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "epoch_capabilities_index",
   "value": 166.09,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 162.89-170.63. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "gpt-6-luna",
   "test": "epoch_capabilities_index",
   "value": 156.28,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 153.67-158.75. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "epoch_capabilities_index",
   "value": 159.62,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 157.17-162.65. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "gpt-oss-120b",
   "test": "epoch_capabilities_index",
   "value": 139.93,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 135.23-142.78. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "claude-opus-5-5",
   "test": "epoch_capabilities_index",
   "value": 167.33,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 164.06-171.68. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "epoch_capabilities_index",
   "value": 165.03,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 161.68-169.06. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "claude-fable-5-1",
   "test": "epoch_capabilities_index",
   "value": 164.7,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 161.63-168.81. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "claude-opus-5",
   "test": "epoch_capabilities_index",
   "value": 162.78,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 160.12-166.47. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "epoch_capabilities_index",
   "value": 142.41,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 139.4-144.07. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "epoch_capabilities_index",
   "value": 156.71,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 154.41-160.26. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "epoch_capabilities_index",
   "value": 154.77,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 152.44-157.27. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "epoch_capabilities_index",
   "value": 145.13,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 142.87-146.56. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "gemma-4-31b",
   "test": "epoch_capabilities_index",
   "value": 142.74,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 140.25-145.1. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "muse-spark-1-3",
   "test": "epoch_capabilities_index",
   "value": 156.75,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 154.78-159.34. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "grok-4-7",
   "test": "epoch_capabilities_index",
   "value": 153.53,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 151.84-155.85. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "epoch_capabilities_index",
   "value": 141.35,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 138.56-143.2. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "deepseek-v4-1-flash",
   "test": "epoch_capabilities_index",
   "value": 154.9,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 148.49-157.33. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "epoch_capabilities_index",
   "value": 155.31,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 153.48-157.45. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "qwen3-8-max",
   "test": "epoch_capabilities_index",
   "value": 155.05,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 153.29-157.19. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "qwen3-8-27b",
   "test": "epoch_capabilities_index",
   "value": 149.38,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 147.3-151.13. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "kimi-k3",
   "test": "epoch_capabilities_index",
   "value": 157.45,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 155.15-160.03. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "glm-5-3",
   "test": "epoch_capabilities_index",
   "value": 155.61,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 153.6-157.92. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "minimax-m3",
   "test": "epoch_capabilities_index",
   "value": 146.95,
   "unit": "ECI points",
   "source": "Epoch AI Capabilities Index",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "90% CI 142.59-149.44. Composite fitted on many benchmarks, including third-party results."
  },
  {
   "model": "gpt-6-astra",
   "test": "gpqa_diamond",
   "value": 95.77,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-30",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-astra_max\" (max over 1 reasoning setting(s)); stderr 1.4 pts. Artificial Analysis (independent) measures 96.1% (GPT-6 Astra (Max))."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "gpqa_diamond",
   "value": 95.39,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6.1-sol_max\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "gpt-6-luna",
   "test": "gpqa_diamond",
   "value": 90.47,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-luna_max\" (max over 1 reasoning setting(s)); stderr 1.9 pts."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "gpqa_diamond",
   "value": 93.31,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-09",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-5.6-terra_max\" (max over 3 reasoning setting(s)); stderr 1.5 pts. Artificial Analysis (independent) measures 92.5% (GPT-5.6 Terra (Max))."
  },
  {
   "model": "gpt-oss-120b",
   "test": "gpqa_diamond",
   "value": 75.76,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2025-12-11",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"openai/gpt-oss-120b_high\" (max over 1 reasoning setting(s)); stderr 2.7 pts. Artificial Analysis (independent) measures 78.2% (gpt-oss-120b (High))."
  },
  {
   "model": "claude-opus-5-5",
   "test": "gpqa_diamond",
   "value": 90.59,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5-5_max\" (max over 1 reasoning setting(s)); stderr 1.9 pts."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "gpqa_diamond",
   "value": 95.58,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-sonnet-5-5_max\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "claude-opus-5",
   "test": "gpqa_diamond",
   "value": 93.88,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-24",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5_max\" (max over 3 reasoning setting(s)); stderr 1.5 pts. Artificial Analysis (independent) measures 93.2% (Claude Opus 5 (Max))."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "gpqa_diamond",
   "value": 71.21,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2025-10-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-haiku-4-5-20251001_32K\" (max over 2 reasoning setting(s)); stderr 3.2 pts. Artificial Analysis (independent) measures 67.2% (Claude 4.5 Haiku (Reasoning))."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "gpqa_diamond",
   "value": 95.39,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.8-flash_high\" (max over 1 reasoning setting(s)); stderr 1.4 pts. Artificial Analysis (independent) measures 95.3% (Gemini 3.8 Flash (High))."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "gpqa_diamond",
   "value": 94.44,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-06",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.1-pro-preview_high\" (max over 2 reasoning setting(s)); stderr 1.6 pts. Artificial Analysis (independent) measures 94.1% (Gemini 3.1 Pro Preview)."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "gpqa_diamond",
   "value": 83.33,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-06",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.5-flash-lite_high\" (max over 3 reasoning setting(s)); stderr 2.7 pts. Artificial Analysis (independent) measures 83.8% (Gemini 3.5 Flash-Lite)."
  },
  {
   "model": "gemma-4-31b",
   "test": "gpqa_diamond",
   "value": 75.76,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-06",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemma-4-31b-it_minimal\" (max over 1 reasoning setting(s)); stderr 3.1 pts. Artificial Analysis (independent) measures 85.7% (Gemma 4 31B (Reasoning))."
  },
  {
   "model": "grok-4-7",
   "test": "gpqa_diamond",
   "value": 92.74,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"grok-4.7_xhigh\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "gpqa_diamond",
   "value": 91.67,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-18",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"deepseek-v4-pro-0813_max\" (max over 1 reasoning setting(s)); stderr 1.5 pts. Artificial Analysis (independent) measures 92.8% (DeepSeek V4 Pro 0813 (Max))."
  },
  {
   "model": "qwen3-8-max",
   "test": "gpqa_diamond",
   "value": 92.3,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"qwen3.8-max-0902_xhigh\" (max over 1 reasoning setting(s)); stderr 1.7 pts. Artificial Analysis (independent) measures 92.8% (Qwen3.8 Max (0902))."
  },
  {
   "model": "kimi-k3",
   "test": "gpqa_diamond",
   "value": 93.12,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-16",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"kimi-k3_max\" (max over 3 reasoning setting(s)); stderr 1.5 pts. Artificial Analysis (independent) measures 93.5% (Kimi K3 (Max))."
  },
  {
   "model": "glm-5-3",
   "test": "gpqa_diamond",
   "value": 90.91,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-24",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"glm-5.3_max\" (max over 1 reasoning setting(s)); stderr 1.6 pts. Artificial Analysis (independent) measures 91.7% (GLM-5.3 (Max))."
  },
  {
   "model": "minimax-m3",
   "test": "gpqa_diamond",
   "value": 90.91,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-10",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"MiniMax-M3\" (max over 2 reasoning setting(s)); stderr 2.0 pts. Artificial Analysis (independent) measures 92.9% (MiniMax-M3)."
  },
  {
   "model": "gpt-6-astra",
   "test": "frontiermath_t1_3",
   "value": 93.68,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-30",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-astra_max\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "frontiermath_t1_3",
   "value": 93.68,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6.1-sol_max\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "gpt-6-luna",
   "test": "frontiermath_t1_3",
   "value": 78.95,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-luna_max\" (max over 1 reasoning setting(s)); stderr 2.4 pts."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "frontiermath_t1_3",
   "value": 85.96,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-09",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-5.6-terra_max\" (max over 1 reasoning setting(s)); stderr 2.1 pts."
  },
  {
   "model": "claude-opus-5-5",
   "test": "frontiermath_t1_3",
   "value": 91.23,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5-5_max\" (max over 1 reasoning setting(s)); stderr 1.7 pts."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "frontiermath_t1_3",
   "value": 88.77,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-sonnet-5-5_max\" (max over 1 reasoning setting(s)); stderr 1.9 pts."
  },
  {
   "model": "claude-fable-5-1",
   "test": "frontiermath_t1_3",
   "value": 90.18,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-fable-5-1_max\" (max over 1 reasoning setting(s)); stderr 1.8 pts."
  },
  {
   "model": "claude-opus-5",
   "test": "frontiermath_t1_3",
   "value": 85.61,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-24",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5_max\" (max over 1 reasoning setting(s)); stderr 2.1 pts."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "frontiermath_t1_3",
   "value": 68.42,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.8-flash_high\" (max over 1 reasoning setting(s)); stderr 2.8 pts."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "frontiermath_t1_3",
   "value": 59.65,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-06-11",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.1-pro-preview\" (max over 1 reasoning setting(s)); stderr 2.9 pts."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "frontiermath_t1_3",
   "value": 25.96,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.5-flash-lite_high\" (max over 1 reasoning setting(s)); stderr 2.6 pts."
  },
  {
   "model": "muse-spark-1-3",
   "test": "frontiermath_t1_3",
   "value": 74.39,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-16",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"muse-spark-1.3_xhigh\" (max over 2 reasoning setting(s)); stderr 2.6 pts."
  },
  {
   "model": "grok-4-7",
   "test": "frontiermath_t1_3",
   "value": 52.98,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"grok-4.7_xhigh\" (max over 1 reasoning setting(s)); stderr 3.0 pts."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "frontiermath_t1_3",
   "value": 64.56,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-19",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"deepseek-v4-pro-0813_max\" (max over 1 reasoning setting(s)); stderr 2.8 pts."
  },
  {
   "model": "qwen3-8-max",
   "test": "frontiermath_t1_3",
   "value": 65.61,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"qwen3.8-max-0902_xhigh\" (max over 1 reasoning setting(s)); stderr 2.8 pts."
  },
  {
   "model": "kimi-k3",
   "test": "frontiermath_t1_3",
   "value": 72.18,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-17",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"kimi-k3_max\" (max over 1 reasoning setting(s)); stderr 2.7 pts."
  },
  {
   "model": "glm-5-3",
   "test": "frontiermath_t1_3",
   "value": 68.77,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-25",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"glm-5.3_max\" (max over 1 reasoning setting(s)); stderr 2.7 pts."
  },
  {
   "model": "gpt-6-astra",
   "test": "frontiermath_t4",
   "value": 97.6,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-30",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-astra_high\" (max over 6 reasoning setting(s)); stderr 2.4 pts."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "frontiermath_t4",
   "value": 100.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6.1-sol_max\" (max over 1 reasoning setting(s)); stderr 0.0 pts."
  },
  {
   "model": "gpt-6-luna",
   "test": "frontiermath_t4",
   "value": 56.1,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-luna_max\" (max over 1 reasoning setting(s)); stderr 7.8 pts."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "frontiermath_t4",
   "value": 70.73,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-09",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-5.6-terra_max\" (max over 1 reasoning setting(s)); stderr 7.2 pts."
  },
  {
   "model": "claude-opus-5-5",
   "test": "frontiermath_t4",
   "value": 95.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5-5_max\" (max over 1 reasoning setting(s)); stderr 4.1 pts."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "frontiermath_t4",
   "value": 80.49,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-sonnet-5-5_max\" (max over 1 reasoning setting(s)); stderr 6.3 pts."
  },
  {
   "model": "claude-fable-5-1",
   "test": "frontiermath_t4",
   "value": 87.8,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-fable-5-1_max\" (max over 1 reasoning setting(s)); stderr 5.2 pts."
  },
  {
   "model": "claude-opus-5",
   "test": "frontiermath_t4",
   "value": 73.17,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-24",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5_max\" (max over 1 reasoning setting(s)); stderr 7.0 pts."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "frontiermath_t4",
   "value": 21.95,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.8-flash_high\" (max over 1 reasoning setting(s)); stderr 6.5 pts."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "frontiermath_t4",
   "value": 26.83,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-06-11",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.1-pro-preview\" (max over 1 reasoning setting(s)); stderr 7.0 pts."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "frontiermath_t4",
   "value": 0.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.5-flash-lite_high\" (max over 1 reasoning setting(s)); stderr 0.0 pts."
  },
  {
   "model": "muse-spark-1-3",
   "test": "frontiermath_t4",
   "value": 46.34,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-18",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"muse-spark-1.3_max\" (max over 2 reasoning setting(s)); stderr 7.9 pts."
  },
  {
   "model": "grok-4-7",
   "test": "frontiermath_t4",
   "value": 17.07,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"grok-4.7_xhigh\" (max over 1 reasoning setting(s)); stderr 5.9 pts."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "frontiermath_t4",
   "value": 26.83,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-19",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"deepseek-v4-pro-0813_max\" (max over 1 reasoning setting(s)); stderr 7.0 pts."
  },
  {
   "model": "qwen3-8-max",
   "test": "frontiermath_t4",
   "value": 34.15,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"qwen3.8-max-0902_xhigh\" (max over 1 reasoning setting(s)); stderr 7.5 pts."
  },
  {
   "model": "kimi-k3",
   "test": "frontiermath_t4",
   "value": 39.02,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-17",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"kimi-k3_max\" (max over 1 reasoning setting(s)); stderr 7.7 pts."
  },
  {
   "model": "glm-5-3",
   "test": "frontiermath_t4",
   "value": 29.27,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-25",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"glm-5.3_max\" (max over 1 reasoning setting(s)); stderr 7.2 pts."
  },
  {
   "model": "gpt-6-astra",
   "test": "otis_mock_aime",
   "value": 100.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-30",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-astra_max\" (max over 1 reasoning setting(s)); stderr 0.0 pts."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "otis_mock_aime",
   "value": 100.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6.1-sol_max\" (max over 1 reasoning setting(s)); stderr 0.0 pts."
  },
  {
   "model": "gpt-6-luna",
   "test": "otis_mock_aime",
   "value": 98.89,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-luna_max\" (max over 1 reasoning setting(s)); stderr 0.7 pts."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "otis_mock_aime",
   "value": 99.72,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-09",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-5.6-terra_max\" (max over 3 reasoning setting(s)); stderr 0.3 pts."
  },
  {
   "model": "gpt-oss-120b",
   "test": "otis_mock_aime",
   "value": 88.89,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2025-12-11",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"openai/gpt-oss-120b_high\" (max over 1 reasoning setting(s)); stderr 4.5 pts."
  },
  {
   "model": "claude-opus-5-5",
   "test": "otis_mock_aime",
   "value": 100.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5-5_max\" (max over 1 reasoning setting(s)); stderr 0.0 pts."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "otis_mock_aime",
   "value": 100.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-sonnet-5-5_max\" (max over 1 reasoning setting(s)); stderr 0.0 pts."
  },
  {
   "model": "claude-fable-5-1",
   "test": "otis_mock_aime",
   "value": 100.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-fable-5-1_max\" (max over 1 reasoning setting(s)); stderr 0.0 pts."
  },
  {
   "model": "claude-opus-5",
   "test": "otis_mock_aime",
   "value": 98.89,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-24",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5_max\" (max over 3 reasoning setting(s)); stderr 1.1 pts."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "otis_mock_aime",
   "value": 66.67,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2025-10-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-haiku-4-5-20251001_32K\" (max over 2 reasoning setting(s)); stderr 7.1 pts."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "otis_mock_aime",
   "value": 98.89,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.8-flash_high\" (max over 1 reasoning setting(s)); stderr 0.9 pts."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "otis_mock_aime",
   "value": 95.6,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-02-20",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.1-pro-preview\" (max over 2 reasoning setting(s)); stderr 3.1 pts."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "otis_mock_aime",
   "value": 71.11,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-06",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.5-flash-lite_high\" (max over 3 reasoning setting(s)); stderr 6.8 pts."
  },
  {
   "model": "gemma-4-31b",
   "test": "otis_mock_aime",
   "value": 73.33,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-06",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemma-4-31b-it_minimal\" (max over 1 reasoning setting(s)); stderr 6.7 pts."
  },
  {
   "model": "muse-spark-1-3",
   "test": "otis_mock_aime",
   "value": 99.17,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-17",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"muse-spark-1.3_xhigh\" (max over 1 reasoning setting(s)); stderr 0.6 pts."
  },
  {
   "model": "grok-4-7",
   "test": "otis_mock_aime",
   "value": 98.06,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"grok-4.7_xhigh\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "otis_mock_aime",
   "value": 98.61,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-18",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"deepseek-v4-pro-0813_max\" (max over 1 reasoning setting(s)); stderr 1.0 pts."
  },
  {
   "model": "qwen3-8-max",
   "test": "otis_mock_aime",
   "value": 100.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"qwen3.8-max-0902_xhigh\" (max over 1 reasoning setting(s)); stderr 0.0 pts."
  },
  {
   "model": "kimi-k3",
   "test": "otis_mock_aime",
   "value": 97.22,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-07-16",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"kimi-k3_max\" (max over 3 reasoning setting(s)); stderr 1.1 pts."
  },
  {
   "model": "glm-5-3",
   "test": "otis_mock_aime",
   "value": 91.11,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-24",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"glm-5.3_max\" (max over 1 reasoning setting(s)); stderr 3.5 pts."
  },
  {
   "model": "minimax-m3",
   "test": "otis_mock_aime",
   "value": 71.11,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-10",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"MiniMax-M3\" (max over 2 reasoning setting(s)); stderr 6.8 pts."
  },
  {
   "model": "gpt-6-astra",
   "test": "simpleqa_verified",
   "value": 75.6,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-30",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-astra_max\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "simpleqa_verified",
   "value": 73.9,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6.1-sol_max\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "gpt-6-luna",
   "test": "simpleqa_verified",
   "value": 41.4,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-6-luna_max\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "simpleqa_verified",
   "value": 43.2,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-10",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gpt-5.6-terra_max\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "claude-opus-5-5",
   "test": "simpleqa_verified",
   "value": 72.2,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5-5_max\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "simpleqa_verified",
   "value": 46.5,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-29",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-sonnet-5-5_max\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "claude-fable-5-1",
   "test": "simpleqa_verified",
   "value": 70.8,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-fable-5-1_max\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "claude-opus-5",
   "test": "simpleqa_verified",
   "value": 59.9,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-10",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-opus-5_max\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "simpleqa_verified",
   "value": 13.2,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-10",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"claude-haiku-4-5-20251001\" (max over 2 reasoning setting(s)); stderr 1.1 pts."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "simpleqa_verified",
   "value": 69.7,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.8-flash_high\" (max over 1 reasoning setting(s)); stderr 1.5 pts."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "simpleqa_verified",
   "value": 73.5,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-10",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemini-3.1-pro-preview_high\" (max over 1 reasoning setting(s)); stderr 1.4 pts."
  },
  {
   "model": "gemma-4-31b",
   "test": "simpleqa_verified",
   "value": 10.4,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-27",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"gemma-4-31b-it\" (max over 1 reasoning setting(s)); stderr 1.0 pts."
  },
  {
   "model": "grok-4-7",
   "test": "simpleqa_verified",
   "value": 56.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-22",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"grok-4.7_xhigh\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "simpleqa_verified",
   "value": 52.91,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-27",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"deepseek-v4-pro-0813_max\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "qwen3-8-max",
   "test": "simpleqa_verified",
   "value": 47.3,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-09-02",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"qwen3.8-max-0902_xhigh\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "kimi-k3",
   "test": "simpleqa_verified",
   "value": 50.6,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-27",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"kimi-k3_max\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "glm-5-3",
   "test": "simpleqa_verified",
   "value": 41.0,
   "unit": "% accuracy",
   "source": "Epoch AI Benchmarking Hub (data export)",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-08-28",
   "kind": "independent",
   "note": "Run by Epoch AI; best variant \"glm-5.3_max\" (max over 1 reasoning setting(s)); stderr 1.6 pts."
  },
  {
   "model": "claude-fable-5-1",
   "test": "gpqa_diamond",
   "value": 93.74,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Claude Fable 5.1 (Max, Default Fallback). No Epoch AI run found."
  },
  {
   "model": "muse-spark-1-3",
   "test": "gpqa_diamond",
   "value": 93.54,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Muse Spark 1.3 (Max). No Epoch AI run found."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "gpqa_diamond",
   "value": 74.85,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Mistral Medium 3.5. No Epoch AI run found."
  },
  {
   "model": "qwen3-8-27b",
   "test": "gpqa_diamond",
   "value": 90.51,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Configuration: Qwen3.8 27B (Xhigh). No Epoch AI run found."
  },
  {
   "model": "claude-haiku-5-5",
   "test": "aa_intelligence_index",
   "value": 43.4,
   "unit": "index points (v4.3.2)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-08",
   "kind": "independent",
   "note": "Configuration: Claude Haiku 5.5 (Max). Artificial Analysis also lists Xhigh 41.25, High 37.82, Medium 34.46, Low 29.45."
  },
  {
   "model": "claude-haiku-5-5",
   "test": "hle",
   "value": 44.39,
   "unit": "% accuracy",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-08",
   "kind": "independent",
   "note": "Configuration: Claude Haiku 5.5 (Max), without tools. Vendor (Anthropic): 45.9% without tools, 57.4% with tools (https://www.anthropic.com/claude-haiku-5-5, read 2026-10-08)."
  },
  {
   "model": "claude-haiku-5-5",
   "test": "terminal_bench_4",
   "value": 32.83,
   "unit": "% tasks passed (pass@1)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-08",
   "kind": "independent",
   "note": "Configuration: Claude Haiku 5.5 (Max). Vendor (Anthropic): 39.2% (https://www.anthropic.com/claude-haiku-5-5, read 2026-10-08)."
  },
  {
   "model": "claude-haiku-5-5",
   "test": "aa_omniscience",
   "value": 10.67,
   "unit": "index points (-100 to 100)",
   "source": "Artificial Analysis",
   "url": "https://artificialanalysis.ai/leaderboards/models",
   "date": "2026-10-08",
   "kind": "independent",
   "note": "Configuration: Claude Haiku 5.5 (Max)."
  },
  {
   "model": "gpt-6-astra",
   "test": "arc_agi_2",
   "value": 95.0,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"GPT-6 Astra (Max)\" (semi-private set), $1.12/task."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "arc_agi_2",
   "value": 94.17,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"GPT-6.1 Sol (Max)\" (semi-private set), $0.25/task."
  },
  {
   "model": "gpt-6-luna",
   "test": "arc_agi_2",
   "value": 59.31,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"GPT-6 Luna (Max)\" (semi-private set), $0.06/task."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "arc_agi_2",
   "value": 83.9,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"GPT-5.6 Terra (Max)\" (semi-private set), $1.09/task."
  },
  {
   "model": "claude-opus-5-5",
   "test": "arc_agi_2",
   "value": 93.33,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Claude Opus 5.5 (High)\" (semi-private set), $0.41/task."
  },
  {
   "model": "claude-fable-5-1",
   "test": "arc_agi_2",
   "value": 90.0,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Claude Fable 5.1 (Max)\" (semi-private set), $4.49/task."
  },
  {
   "model": "claude-opus-5",
   "test": "arc_agi_2",
   "value": 90.42,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Claude Opus 5 (Max)\" (semi-private set), $2.06/task."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "arc_agi_2",
   "value": 4.03,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Claude Haiku 4.5 (Thinking 32K)\" (semi-private set), $0.38/task."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "arc_agi_2",
   "value": 89.17,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Gemini 3.8 Flash (High)\" (semi-private set), $0.4/task."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "arc_agi_2",
   "value": 77.08,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Gemini 3.1 Pro (Preview)\" (semi-private set), $0.96/task."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "arc_agi_2",
   "value": 10.28,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Gemini 3.5 Flash-Lite (High)\" (semi-private set), $0.14/task."
  },
  {
   "model": "grok-4-7",
   "test": "arc_agi_2",
   "value": 61.39,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Grok 4.7 (XHigh)\" (semi-private set), $2.01/task."
  },
  {
   "model": "deepseek-v4-1-flash",
   "test": "arc_agi_2",
   "value": 72.92,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"DeepSeek V4.1 Flash (Max)\" (semi-private set), $0.13/task."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "arc_agi_2",
   "value": 61.25,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"DeepSeek V4 Pro 0813 (Max)\" (semi-private set), $0.6/task."
  },
  {
   "model": "qwen3-8-27b",
   "test": "arc_agi_2",
   "value": 42.36,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Qwen3.8-27B (XHigh)\" (semi-private set), $0.45/task."
  },
  {
   "model": "kimi-k3",
   "test": "arc_agi_2",
   "value": 60.42,
   "unit": "% of tasks solved",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Kimi K3 (Max)\" (semi-private set), $1.59/task."
  },
  {
   "model": "gpt-6-astra",
   "test": "arc_agi_3",
   "value": 62.71,
   "unit": "% score (100% = beats every game as efficiently as humans)",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"GPT-6 Astra (Max)\" (semi-private set). With the \"Provider Adapter\" harness ARC Prize also lists 99.9% (GPT-6 Astra - Provider Adapter (High)); not used here because the harness differs."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "arc_agi_3",
   "value": 52.73,
   "unit": "% score (100% = beats every game as efficiently as humans)",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"GPT-6.1 Sol (Max)\" (semi-private set). With the \"Provider Adapter\" harness ARC Prize also lists 96.4% (GPT-6.1 Sol - Provider Adapter (XHigh)); not used here because the harness differs."
  },
  {
   "model": "gpt-6-luna",
   "test": "arc_agi_3",
   "value": 0.19,
   "unit": "% score (100% = beats every game as efficiently as humans)",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"GPT-6 Luna (Medium)\" (semi-private set). With the \"Provider Adapter\" harness ARC Prize also lists 0.6% (GPT-6 Luna - Provider Adapter (Max)); not used here because the harness differs."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "arc_agi_3",
   "value": 0.8,
   "unit": "% score (100% = beats every game as efficiently as humans)",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"GPT-5.6 Terra (Max)\" (semi-private set)."
  },
  {
   "model": "claude-opus-5",
   "test": "arc_agi_3",
   "value": 30.16,
   "unit": "% score (100% = beats every game as efficiently as humans)",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Claude Opus 5 (High)\" (semi-private set)."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "arc_agi_3",
   "value": 10.37,
   "unit": "% score (100% = beats every game as efficiently as humans)",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Gemini 3.8 Flash (High)\" (semi-private set). With the \"Provider Adapter\" harness ARC Prize also lists 35.0% (Gemini 3.8 Flash - Provider Adapter (High)); not used here because the harness differs."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "arc_agi_3",
   "value": 0.42,
   "unit": "% score (100% = beats every game as efficiently as humans)",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Gemini 3.1 Pro (Preview)\" (semi-private set)."
  },
  {
   "model": "grok-4-7",
   "test": "arc_agi_3",
   "value": 1.82,
   "unit": "% score (100% = beats every game as efficiently as humans)",
   "source": "ARC Prize verified leaderboard",
   "url": "https://arcprize.org/leaderboard",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Best verified setting \"Grok 4.7 (XHigh)\" (semi-private set). With the \"Provider Adapter\" harness ARC Prize also lists 10.0% (Grok 4.7 - Provider Adapter (High)); not used here because the harness differs."
  },
  {
   "model": "claude-fable-5-1",
   "test": "mmlu_pro",
   "value": 92.38,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 1 on Vals.ai (\"Claude Fable 5.1\")."
  },
  {
   "model": "claude-opus-5",
   "test": "mmlu_pro",
   "value": 91.59,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 2 on Vals.ai (\"Claude Opus 5\")."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "mmlu_pro",
   "value": 90.99,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 4 on Vals.ai (\"Gemini 3.1 Pro Preview (02/26)\")."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "mmlu_pro",
   "value": 90.22,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 5 on Vals.ai (\"Gemini 3.8 Flash\")."
  },
  {
   "model": "qwen3-8-max",
   "test": "mmlu_pro",
   "value": 88.6,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 18 on Vals.ai (\"Qwen 3.8 Max\"). Vals label \"Qwen 3.8 Max\" may be the 2026-08 version rather than the 0902 refresh."
  },
  {
   "model": "kimi-k3",
   "test": "mmlu_pro",
   "value": 87.97,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 22 on Vals.ai (\"Kimi K3\")."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "mmlu_pro",
   "value": 86.97,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 36 on Vals.ai (\"DeepSeek V4 Pro 0813\")."
  },
  {
   "model": "glm-5-3",
   "test": "mmlu_pro",
   "value": 86.77,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 38 on Vals.ai (\"GLM 5.3\")."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "mmlu_pro",
   "value": 86.66,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 40 on Vals.ai (\"GPT-5.6 Terra\")."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "mmlu_pro",
   "value": 85.84,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 54 on Vals.ai (\"Gemini 3.5 Flash Lite\")."
  },
  {
   "model": "qwen3-8-27b",
   "test": "mmlu_pro",
   "value": 84.34,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 65 on Vals.ai (\"Qwen 3.8 27B\")."
  },
  {
   "model": "minimax-m3",
   "test": "mmlu_pro",
   "value": 84.22,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 66 on Vals.ai (\"MiniMax-M3\")."
  },
  {
   "model": "gpt-oss-120b",
   "test": "mmlu_pro",
   "value": 79.17,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 100 on Vals.ai (\"GPT OSS 120B\")."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "mmlu_pro",
   "value": 78.72,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 102 on Vals.ai (\"Claude Haiku 4.5 (Thinking)\")."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "mmlu_pro",
   "value": 75.33,
   "unit": "% accuracy",
   "source": "Vals.ai MMLU Pro",
   "url": "https://www.vals.ai/benchmarks/mmlu_pro",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 111 on Vals.ai (\"Mistral Medium 3.5\")."
  },
  {
   "model": "claude-fable-5-1",
   "test": "livecodebench",
   "value": 90.52,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 1 on Vals.ai (\"Claude Fable 5.1\")."
  },
  {
   "model": "claude-opus-5",
   "test": "livecodebench",
   "value": 89.03,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 4 on Vals.ai (\"Claude Opus 5\")."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "livecodebench",
   "value": 88.48,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 6 on Vals.ai (\"Gemini 3.1 Pro Preview (02/26)\")."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "livecodebench",
   "value": 89.48,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 3 on Vals.ai (\"Gemini 3.8 Flash\")."
  },
  {
   "model": "qwen3-8-max",
   "test": "livecodebench",
   "value": 87.85,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 10 on Vals.ai (\"Qwen 3.8 Max\"). Vals label \"Qwen 3.8 Max\" may be the 2026-08 version rather than the 0902 refresh."
  },
  {
   "model": "kimi-k3",
   "test": "livecodebench",
   "value": 87.19,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 18 on Vals.ai (\"Kimi K3\")."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "livecodebench",
   "value": 87.53,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 13 on Vals.ai (\"DeepSeek V4 Pro 0813\")."
  },
  {
   "model": "glm-5-3",
   "test": "livecodebench",
   "value": 80.53,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 68 on Vals.ai (\"GLM 5.3\")."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "livecodebench",
   "value": 85.93,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 27 on Vals.ai (\"GPT-5.6 Terra\")."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "livecodebench",
   "value": 79.01,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 75 on Vals.ai (\"Gemini 3.5 Flash Lite\")."
  },
  {
   "model": "qwen3-8-27b",
   "test": "livecodebench",
   "value": 84.0,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 43 on Vals.ai (\"Qwen 3.8 27B\")."
  },
  {
   "model": "minimax-m3",
   "test": "livecodebench",
   "value": 82.15,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 56 on Vals.ai (\"MiniMax-M3\")."
  },
  {
   "model": "gpt-oss-120b",
   "test": "livecodebench",
   "value": 83.23,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 51 on Vals.ai (\"GPT OSS 120B\")."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "livecodebench",
   "value": 41.17,
   "unit": "% accuracy",
   "source": "Vals.ai LiveCodeBench",
   "url": "https://www.vals.ai/benchmarks/lcb",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 130 on Vals.ai (\"Claude Haiku 4.5 (Thinking)\")."
  },
  {
   "model": "claude-opus-5",
   "test": "swe_bench_verified",
   "value": 97.0,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 1 on Vals.ai (\"Claude Opus 5\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "swe_bench_verified",
   "value": 78.8,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 29 on Vals.ai (\"Gemini 3.1 Pro Preview (02/26)\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "swe_bench_verified",
   "value": 80.0,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 24 on Vals.ai (\"Gemini 3.8 Flash\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "qwen3-8-max",
   "test": "swe_bench_verified",
   "value": 85.6,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 17 on Vals.ai (\"Qwen 3.8 Max\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness. Vals label \"Qwen 3.8 Max\" may be the 2026-08 version rather than the 0902 refresh."
  },
  {
   "model": "kimi-k3",
   "test": "swe_bench_verified",
   "value": 93.4,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 8 on Vals.ai (\"Kimi K3\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "swe_bench_verified",
   "value": 96.4,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 2 on Vals.ai (\"DeepSeek V4 Pro 0813\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "glm-5-3",
   "test": "swe_bench_verified",
   "value": 95.4,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 6 on Vals.ai (\"GLM 5.3\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "swe_bench_verified",
   "value": 95.4,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 5 on Vals.ai (\"GPT-5.6 Terra\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "swe_bench_verified",
   "value": 75.0,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 45 on Vals.ai (\"Gemini 3.5 Flash Lite\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "qwen3-8-27b",
   "test": "swe_bench_verified",
   "value": 86.0,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 15 on Vals.ai (\"Qwen 3.8 27B\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "minimax-m3",
   "test": "swe_bench_verified",
   "value": 75.0,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 46 on Vals.ai (\"MiniMax-M3\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "gpt-oss-120b",
   "test": "swe_bench_verified",
   "value": 33.6,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 87 on Vals.ai (\"GPT OSS 120B\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "swe_bench_verified",
   "value": 66.6,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 71 on Vals.ai (\"Claude Haiku 4.5 (Thinking)\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "mistral-medium-3-5",
   "test": "swe_bench_verified",
   "value": 66.4,
   "unit": "% issues resolved",
   "source": "Vals.ai SWE-bench Verified",
   "url": "https://www.vals.ai/benchmarks/swebench",
   "date": "2026-09-01",
   "kind": "independent",
   "note": "Rank 72 on Vals.ai (\"Mistral Medium 3.5\"). Harness: Mini-SWE-agent; Vals runs all models with the same harness."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "swe_bench_pro",
   "value": 46.1,
   "unit": "% resolved (public set)",
   "source": "Scale AI SEAL SWE-Bench Pro (public set)",
   "url": "https://scale.com/leaderboard/swe_bench_pro_public",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"gemini-3.1-pro (thinking)*\", ±3.6. Leaderboard not updated for models released after mid-2026."
  },
  {
   "model": "claude-haiku-4-5",
   "test": "swe_bench_pro",
   "value": 39.45,
   "unit": "% resolved (public set)",
   "source": "Scale AI SEAL SWE-Bench Pro (public set)",
   "url": "https://scale.com/leaderboard/swe_bench_pro_public",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"claude-4-5-haiku\", ±3.55. Leaderboard not updated for models released after mid-2026."
  },
  {
   "model": "gpt-oss-120b",
   "test": "swe_bench_pro",
   "value": 16.2,
   "unit": "% resolved (public set)",
   "source": "Scale AI SEAL SWE-Bench Pro (public set)",
   "url": "https://scale.com/leaderboard/swe_bench_pro_public",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"gpt-oss-120b\", ±2.67. Leaderboard not updated for models released after mid-2026."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "metr_time_horizon_50",
   "value": 384.15,
   "unit": "minutes (50% time horizon)",
   "source": "METR time horizons (Time Horizon 1.1)",
   "url": "https://metr.org/assets/benchmark_results_1_1.yaml",
   "date": "2026-04-15",
   "kind": "independent",
   "note": "95% CI 234-695 min; 80% horizon 89.8 min. Added by METR on 2026-04-15."
  },
  {
   "model": "gpt-oss-120b",
   "test": "metr_time_horizon_50",
   "value": 41.98,
   "unit": "minutes (50% time horizon)",
   "source": "METR time horizons (Time Horizon 1.1)",
   "url": "https://metr.org/blog/2025-03-19-measuring-ai-ability-to-complete-long-tasks/",
   "date": "2025-08",
   "kind": "independent",
   "note": "95% CI 19-82 min. Measured on METR's earlier task suite (v1.0), as compiled by Epoch AI; not in the current Time Horizon 1.1 file."
  },
  {
   "model": "claude-opus-5",
   "test": "osworld_2",
   "value": 31.43,
   "unit": "% tasks fully completed (binary, 500 steps)",
   "source": "OSWorld 2.0 leaderboard (XLANG Lab)",
   "url": "https://osworld-v2.xlang.ai/",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Max effort, batched tool calls, benchmark release v2026.08.08. On the newer v2.1 release the same model scores 44.33% (not comparable across releases)."
  },
  {
   "model": "minimax-m3",
   "test": "osworld_2",
   "value": 4.6,
   "unit": "% tasks fully completed (binary, 500 steps)",
   "source": "OSWorld 2.0 leaderboard (XLANG Lab)",
   "url": "https://osworld-v2.xlang.ai/",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Release v2026.06.24."
  },
  {
   "model": "claude-opus-5",
   "test": "osworld_2_partial",
   "value": 68.31,
   "unit": "% partial score",
   "source": "OSWorld 2.0 leaderboard (XLANG Lab)",
   "url": "https://osworld-v2.xlang.ai/",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Max effort, v2026.08.08. Anthropic reports 74.0% partial on OSWorld 2.1 and 75.4% partial on OSWorld 2.0 (vendor)."
  },
  {
   "model": "minimax-m3",
   "test": "osworld_2_partial",
   "value": 22.3,
   "unit": "% partial score",
   "source": "OSWorld 2.0 leaderboard (XLANG Lab)",
   "url": "https://osworld-v2.xlang.ai/",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Release v2026.06.24."
  },
  {
   "model": "claude-opus-5-5",
   "test": "osworld_2_partial",
   "value": 81.8,
   "unit": "% partial score",
   "source": "Anthropic, Introducing Claude Opus 5.5",
   "url": "https://www.anthropic.com/claude-opus-5-5",
   "date": "2026-09-22",
   "kind": "vendor",
   "note": "OSWorld 2.1 (newer release than the public 2.0 leaderboard), vendor-run. Not comparable with the independent 2.0 numbers."
  },
  {
   "model": "claude-fable-5-1",
   "test": "osworld_2_partial",
   "value": 77.9,
   "unit": "% partial score",
   "source": "Anthropic, Introducing Claude Fable 5.1 and Claude Mythos 5.1",
   "url": "https://www.anthropic.com/claude-fable-and-mythos-5-1",
   "date": "2026-09-01",
   "kind": "vendor",
   "note": "OSWorld 2.0, vendor-run; strict (binary) score reported as 41.7%. Anthropic later reports 80.7% partial on OSWorld 2.1."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "osworld_2_partial",
   "value": 80.1,
   "unit": "% partial score",
   "source": "Anthropic, Introducing Claude Sonnet 5.5",
   "url": "https://www.anthropic.com/claude-sonnet-5-5",
   "date": "2026-09-28",
   "kind": "vendor",
   "note": "OSWorld 2.1, vendor-run."
  },
  {
   "model": "gemini-4-argon",
   "test": "osworld_2_partial",
   "value": 69.2,
   "unit": "% partial score",
   "source": "Google DeepMind, Gemini 4 Argon model page",
   "url": "https://deepmind.google/models/gemini/",
   "date": "2026-09-30",
   "kind": "vendor",
   "note": "\"Offline subset\" of OSWorld-2.0, vendor-run; Google reports GPT-6 Astra at 72.6% on the same subset."
  },
  {
   "model": "gpt-6-astra",
   "test": "osworld_2_partial",
   "value": 72.6,
   "unit": "% partial score",
   "source": "OpenAI, GPT-6 Astra launch page",
   "url": "https://openai.com/index/gpt-6-astra/",
   "date": "2026-10-06",
   "kind": "vendor",
   "note": "OSWorld 2.0 (v2026.08.08), offline set, partial reward; Max effort (Low 61.4, High 71.6, Xhigh 71.7, Max 72.6). OpenAI's later posts on GPT-6 Sol, Luna and GPT-6.1 Sol chart Astra (Max) at 73.5% on the same test. Google DeepMind's Gemini 4 Argon page also quotes 72.6%. Source corrected on 2026-10-06."
  },
  {
   "model": "gpt-6-astra",
   "test": "apex_agents",
   "value": 64.7,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"GPT-6 Astra\"; standard error 5.6 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "apex_agents",
   "value": 60.0,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"GPT-6.1 Sol\"; standard error 5.6 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "gpt-6-luna",
   "test": "apex_agents",
   "value": 44.3,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"GPT-6 Luna\"; standard error 5.5 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "gpt-5-6-terra",
   "test": "apex_agents",
   "value": 58.2,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"GPT-5.6 Terra\"; standard error 5.3 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "gpt-oss-120b",
   "test": "apex_agents",
   "value": 4.4,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"GPT-OSS-120B\"; standard error 2.0 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "claude-opus-5-5",
   "test": "apex_agents",
   "value": 73.5,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Opus 5.5 [claude-opus-5-5]\"; standard error 4.9 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "claude-sonnet-5-5",
   "test": "apex_agents",
   "value": 75.5,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Sonnet 5.5 [claude-sonnet-5-5]\"; standard error 4.7 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "claude-fable-5-1",
   "test": "apex_agents",
   "value": 68.6,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Fable 5.1 [claude-fable-5.1]\"; standard error 4.9 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "claude-opus-5",
   "test": "apex_agents",
   "value": 65.8,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Opus 5\"; standard error 5.1 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "gemini-4-argon",
   "test": "apex_agents",
   "value": 82.2,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Gemini 4 Argon\"; standard error 4.4 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "gemini-3-8-flash",
   "test": "apex_agents",
   "value": 64.3,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Gemini 3.8 Flash\"; standard error 5.3 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "gemini-3-1-pro",
   "test": "apex_agents",
   "value": 35.3,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Gemini 3.1 Pro\"; standard error 5.0 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "gemini-3-5-flash-lite",
   "test": "apex_agents",
   "value": 29.3,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Gemini 3.5 Flash Lite\"; standard error 4.9 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "muse-spark-1-3",
   "test": "apex_agents",
   "value": 57.8,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Muse Spark 1.3\"; standard error 5.5 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "grok-4-7",
   "test": "apex_agents",
   "value": 54.6,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Grok 4.7\"; standard error 5.1 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "deepseek-v4-1-flash",
   "test": "apex_agents",
   "value": 39.5,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"DeepSeek-V4.1-Flash\"; standard error 4.9 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "deepseek-v4-pro-0813",
   "test": "apex_agents",
   "value": 47.3,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"DeepSeek-V4-Pro-0813\"; standard error 5.6 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "qwen3-8-max",
   "test": "apex_agents",
   "value": 63.3,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Qwen3.8-Max\"; standard error 4.9 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "qwen3-8-27b",
   "test": "apex_agents",
   "value": 47.5,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Qwen 3.8 27B\"; standard error 4.9 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "kimi-k3",
   "test": "apex_agents",
   "value": 50.6,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"Kimi K3\"; standard error 5.1 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "glm-5-3",
   "test": "apex_agents",
   "value": 56.6,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"GLM-5.3\"; standard error 5.1 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "minimax-m3",
   "test": "apex_agents",
   "value": 37.7,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"MiniMax-M3\"; standard error 4.9 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "mimo-v2-6-pro",
   "test": "apex_agents",
   "value": 59.5,
   "unit": "% pass@1",
   "source": "APEX-Agents (Mercor) via Epoch AI data export",
   "url": "https://epoch.ai/data/benchmark_data.zip",
   "date": "2026-10-06",
   "kind": "independent",
   "note": "Entry \"MiMo V2.6 Pro\"; standard error 5.2 pts. Run by Mercor (benchmark owner), compiled by Epoch AI."
  },
  {
   "model": "gpt-6-1-sol",
   "test": "osworld_2_partial",
   "value": 71.4,
   "unit": "% partial score",
   "source": "OpenAI, launch page",
   "url": "https://openai.com/index/introducing-gpt-6-1-sol/",
   "date": "2026-10-06",
   "kind": "vendor",
   "note": "OSWorld 2.0 (v2026.08.08), offline set, partial reward; Max effort (Low 59.0, Medium 66.8, High 69.6, Xhigh 69.4, Max 71.4)."
  },
  {
   "model": "gpt-6-luna",
   "test": "osworld_2_partial",
   "value": 52.68,
   "unit": "% partial score",
   "source": "OpenAI, launch page",
   "url": "https://openai.com/index/introducing-gpt-6-sol-and-luna/",
   "date": "2026-10-06",
   "kind": "vendor",
   "note": "OSWorld 2.0 (v2026.08.08), offline set, partial reward; Max effort (Low 8.26, Medium 31.54, High 41.43, Xhigh 46.70, Max 52.68)."
  },
  {
   "model": "claude-haiku-5-5",
   "test": "osworld_2_partial",
   "value": 72.4,
   "unit": "% partial score",
   "source": "Anthropic, Introducing Claude Haiku 5.5",
   "url": "https://www.anthropic.com/claude-haiku-5-5",
   "date": "2026-10-07",
   "kind": "vendor",
   "note": "OSWorld 2.1, 'offline subset', partial-credit score (axis label of Anthropic's chart), vendor-run. Same table: Claude Sonnet 5.5 83.9%, Claude Haiku 4.5 15.7%, GPT-6 Luna 48.9% (offline subset). Not comparable with the independent 2.0 numbers."
  }
 ],
 "gaps": [
  "Claude Haiku 5.5 (released 2026-10-07): only Artificial Analysis results exist on 2026-10-08 (intelligence index, HLE, Terminal-Bench 4.0, AA-Omniscience; no GPQA or tau3-Banking run). Not yet on LMArena, Epoch AI (no ECI, GPQA, FrontierMath, SimpleQA, APEX-Agents), ARC Prize, Vals.ai or OSWorld 2.0. Anthropic reports 72.4% on the OSWorld 2.1 'offline subset' without saying partial or strict; not added to the partial-score table. On the same page Anthropic lists Sonnet 5.5 at 83.9% (offline subset), while the Sonnet 5.5 page reports 80.1% (partial): not the same measure.",
  "LMArena (now arena.ai): no snapshot after 2026-10-02 as of 2026-10-08, so Claude Haiku 5.5 and Mistral Large 4 have no rating yet.",
  "METR 50% time horizon: no published measurement for any model released after April 2026 (GPT-6 Astra, GPT-6.1 Sol, Claude Opus 5.5, Claude Fable 5.1, Gemini 4 Argon, Claude Haiku 5.5, etc.); only Gemini 3.1 Pro and gpt-oss-120b (older suite) are covered. Re-checked 2026-10-08: last entry still dated April 2026.",
  "SWE-bench Verified, MMLU-Pro and LiveCodeBench: Vals.ai stopped running them on new releases (results as of 2026-09-01, pages republished 2026-10-08 unchanged), so GPT-6 Astra, GPT-6.1 Sol, GPT-6 Luna, Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 5.5, Gemini 4 Argon, Grok 4.7, Muse Spark 1.3 and Mistral Large 4 have no independent value. Vendors no longer report SWE-bench Verified.",
  "SWE-Bench Pro: the Scale AI SEAL leaderboard has not added models released after mid-2026 (unchanged on 2026-10-08); only Gemini 3.1 Pro, Claude Haiku 4.5 and gpt-oss-120b are covered in our set.",
  "OSWorld: the OSWorld 2.0 public leaderboard covers only Claude Opus 5 and MiniMax-M3 in our set (unchanged on 2026-10-08); OSWorld-Verified (1.0) has no rows for current models in the Epoch data.",
  "ARC-AGI-3: not yet run for Claude Opus 5.5, Claude Sonnet 5.5, Claude Fable 5.1, Claude Haiku 5.5, Gemini 4 Argon, Kimi K3, DeepSeek, Qwen, GLM, MiniMax, Mistral, Muse Spark (re-checked 2026-10-08 in ARC Prize's evaluations.json).",
  "AIME 2025 / AIME 2026: no independent table covering current models was found (Artificial Analysis retired AIME 2025); OTIS Mock AIME 2024-2025 (Epoch AI) is used instead.",
  "Multilingual / French: Artificial Analysis runs a Multilingual Index (Global-MMLU-Lite, including French) but the per-model values were not exposed in the data retrieved; no French-specific benchmark result was collected.",
  "MRCR / long-context: not collected (AA-LCR is only available inside the AA index; Google reports GraphWalks vendor-only).",
  "tau2-bench (Telecom): retired by Artificial Analysis; tau3-Banking is used, but AA has not run it on several September-October 2026 models (Claude Opus 5.5, Sonnet 5.5, Haiku 5.5, Gemini 4 Argon, GPT-6.1 Sol, Grok 4.7, Mistral Large 4, DeepSeek V4.1 Flash, MiMo-V2.6-Pro).",
  "Gemini 4 Argon, MiMo-V2.6-Pro, Mistral Large 4 and Claude Haiku 5.5 are not yet in the Epoch AI data export of 2026-10-08 (no ECI, GPQA, FrontierMath, SimpleQA values).",
  "GPT-5.6 Terra vendor numbers and GPT-6 Astra / GPT-6.1 Sol vendor tables were not read directly from OpenAI as tables (the GPT-6.1 Sol page renders without its table); OpenAI figures appear only as quoted by Anthropic or Google, or in OpenAI's running text.",
  "Item counts for FrontierMath v2 tiers, OTIS Mock AIME, ARC-AGI-2/3, APEX-Agents and the creator of SimpleQA Verified were not verified in a source read (left null).",
  "Human baselines are not published for most agentic benchmarks (Terminal-Bench, tau3-Banking, APEX-Agents, SWE-bench)."
 ],
 "caveats": [
  "Reasoning effort matters: most 2026 models expose several effort levels (low to max). Epoch AI and ARC Prize values here are the best setting reported; Artificial Analysis values use the highest-effort configuration listed (named in each note). Cost changes a lot between settings (ARC-AGI-2 cost per task for Claude Opus 5.5: $0.24 at low effort vs $1.85 at max).",
  "Harness matters for agentic tests (Terminal-Bench, SWE-bench, OSWorld, ARC-AGI-3): e.g. Gemini 3.1 Pro on Terminal-Bench 2.0 scores 80.2% with TongAgents vs 59.4% with Gemini CLI (tbench.ai via Epoch AI), and GPT-6 Astra scores 62.7% on ARC-AGI-3 with the standard harness vs 98.6% with the Provider Adapter. Only compare values from the same source.",
  "Versions change: Terminal-Bench 2.x vs 4.0, OSWorld 1.0 vs 2.0 vs 2.1, AA Intelligence Index v4.0 to v4.3.2, HLE vs HLE-Rolling. Never compare numbers across versions.",
  "Vendor-reported values (kind = \"vendor\") are run by the model maker with its own settings, often with tools and maximum effort; seven such cells are included (all OSWorld partial scores) and many more vendor figures are kept in notes. Independent values are preferred throughout.",
  "HLE: the independent AA values are without tools; vendors mostly publish with-tools values (4-7 points higher in Anthropic's own no-tools vs with-tools figures).",
  "Saturated tests (GPQA Diamond, OTIS Mock AIME, SWE-bench Verified, MMLU-Pro, LiveCodeBench, ARC-AGI-2 at the top) no longer separate frontier models; gaps of 1-2 points are within noise.",
  "LMArena measures human preference (style, tone) rather than correctness; new models with few votes are marked preliminary and their rank can move.",
  "Composite indices (AA Intelligence Index, Epoch ECI) depend on their component choice and weights; they rank the same top models differently in places (e.g. Claude Sonnet 5.5 is 2nd on AA, 4th on ECI).",
  "Open-weight status follows Artificial Analysis and LMArena licences; Epoch AI classifies GLM-5.3 as closed weights, which conflicts.",
  "Grok 4.7 creator appears as \"SpaceXAI\" on Artificial Analysis and LMArena and as xAI on Epoch AI and ARC Prize.",
  "Dates: \"date\" is the publication or run date of the measurement where available (Epoch run start date, LMArena snapshot 2026-10-02, Vals.ai 2026-09-01); otherwise the date of first retrieval (2026-10-06, or 2026-10-08 for values added on that day). Every value was re-read at its source on 2026-10-08 and none had changed; dates were kept.",
  "Gemini 3.1 Pro (Preview), Artificial Analysis Intelligence Index 29.72 (v4.3.2): confirmed on 2026-10-06 as the only configuration listed. Its knowledge scores match Gemini 3.8 Flash (HLE 47.0% vs 47.8%); the gap comes from the agentic tests (Terminal-Bench 4.0: 4.0% vs 19.7%; GDPval normalized 0.147 vs 0.468), read from the data embedded in https://artificialanalysis.ai/leaderboards/models.",
  "Claude Haiku 5.5 counts more tokens than Haiku 4.5 for the same text (newer tokenizer, about +30% per Anthropic), so per-token prices and per-task costs are not directly comparable across the two generations.",
  "Artificial Analysis lists Mistral Large 4 Preview at its list price ($1.36/$4.18) and a 524,288-token context, whereas Mistral's own page shows the launch price ($0.68/$2.09) and 1M tokens."
 ]
}