Benchmarks
478 benchmarks · 199 in index
Agentic
137 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
τ²-Bench Tool-Agent-User EvaluationVictor Barres et al.150GLM-5.299.1%GDPval-AA normalizedArtificial Analysis134Claude Opus 5.568.3%Terminal-Bench 2.1 · BenchLM86SWE-292.8%Terminal-Bench 2.1Vals AI75Claude Opus 5.587.6%LiveBench Agentic CodingLiveBench66DeepSeek V4.1 Flash 101777.3%Terminal-Bench 2.0Vals AI66GPT-5.573.2%BrowseComp63Atria Dawn Preview92.5%Terminal-Bench 4.0Vals AI61Claude Fable 5.265.2%DeepSWEDatacurve AI59Muse Wonder Full79.0%Agent Arena steerabilityLMArena54Claude Fable 5.233.0Agent Arena task outcomeLMArena54Claude Fable 5.222.0Gert Labs Composite Game BenchmarkGert Labs52Claude Opus 4.873.0%LiveBench DesignLiveBench52Claude Opus 5.586.9%LiveBench ResearchLiveBench52Claude Opus 588.9%MCP Atlas52Muse Spark Open89.2%LiveBench Long AgenticLiveBench51Claude Opus 5.585.0%Agent Arena command recoveryLMArena50Claude Sonnet 5.515.0LiveBench Browser UseLiveBench49GPT-6.1 Terra100.0%MazeBench: visual spatial reasoning in a 3D open worldJonathan Pappas and David Pappas48GPT-6 Astra23.0%Terminal-Bench 1.0Vals AI46GPT-5.263.8%Claw-EvalBowen Ye et al.42Ornith-1.5-397B81.4%Terminal-Bench 2.0 · BenchLM41GPT-5.582.0%OSWorld-VerifiedTianbao Xie et al.37Qwen3.8 Max86.1%Humanity's Last Exam with tools36Kimi K3.1 Pro70.0%JobBenchYuetai Li et al.36Muse Wonder Full75.3%AutomationBench32MiniMax M3.1 Flash62.2%τ³-Bench Tool-Agent-User EvaluationSierra Research32Mercury 2.596.0%APEX-Agents-AAArtificial Analysis / Mercor30Gemini 3.5 Flash47.1%CyberGymZhun Wang et al.29MiMo-V2.6-Flash95.1%τ²-bench BankingSierra Research28PrincipledMachines Cortex 0.173.2%Artificial Analysis AnalystAgentArtificial Analysis26Gemini 3.7 Flash60.0%Berkeley Function Calling Leaderboard v4Shishir G. Patil et al.26BTL-388.5%OSWorld 2.0Mengqi Yuan et al.25Kimi K3.1 Pro79.6%Artificial Analysis ITBench-AAArtificial Analysis24GPT-5.6 Sol56.2%DeepSearchQA24Atria Dawn Preview96.0%Artificial Analysis AutomationBenchArtificial Analysis23Gemini 4 Carbon77.8%ExploitGymZhun Wang et al.23Gemini 4 Carbon53.5%Terminal-Bench 4.0.0Terminal-Bench23Claude Opus 5.564.8%Terminal-Bench 4.1.0Terminal-Bench23Claude Opus 5.565.8%Terminal-Bench HardArtificial Analysis23GPT-5.6 Sol65.9%Toolathlon-Verified23Claude Opus 580.6%τ²-bench RetailSierra Research23Gemini 3.0 Pro85.3%Artificial Analysis GDP.pdfArtificial Analysis22Muse Wonder Full32.6%Artificial Analysis Tau3-BankingArtificial Analysis22Grok 4.650.7%Artificial Analysis Terminal-Bench v4.0Artificial Analysis22Kimi K3.1 Pro67.2%Toolathlon22Muse Spark 1.175.6%τ²-bench AirlineSierra Research22Claude Opus 4.584.0%Agents' Last Exam21GPT-6.1 Terra60.0%Artificial Analysis DeepsecBench-AAArtificial Analysis; dataset by Vercel21GPT-6 Sol46.3%Artificial Analysis Terminal-Bench v2.1Artificial Analysis21Claude Fable 5.191.4%Artificial Analysis CWE-Bench-AAArtificial Analysis; dataset by Collinear AI19Grok 4.768.3%Artificial Analysis Cyber IndexArtificial Analysis with the Cyber Index Alliance (Collinear AI et al.19Kimi K3.1 Pro64.1%Artificial Analysis CyberGym-E2E-AAArtificial Analysis; dataset by Berkeley RDI19Kimi K3.1 Pro94.7%Multimodal Knowledge Agent BenchmarkGoogle DeepMind, in partnership with Artificial Analysis19Claude Fable 5.256.6%ApprenticeBench: end-to-end computer use, continual learning, and long-horizon agency on a real accounts-payable jobNeoCognition18Claude Fable 5.172.0%τ²-bench TelecomSierra Research18Qwen3 Max Thinking98.2%Hermes Index: agentic capability and cost per task inside Hermes AgentNous Research16Claude Opus 5.563.3%WideResearch16Hy4 preview83.9%APEX-Agents13Hy475.2%Terminal-Bench 3.0Ryan Marten et al.13Claude Opus 542.7%VITA-BenchMeituan LongCat Team13Qwen3.7 Max47.9%CWE-bench v1Collinear AI12Gemini 4 Carbon71.0%CWE-bench v2Collinear AI11Claude Opus 5.567.1%OSWorld 2.1 (all 108 tasks)Mengqi Yuan et al.11Claude Opus 5.581.8%Terminal-Bench-Science 0.1Terminal-Bench-Science Team11Claude Mythos 5.275.3%AndroidWorld10Qwen3.8-Omni-Flash87.1%OSWorld 2.1 offline subset (82 tasks)Mengqi Yuan et al.10Claude Fable 5.289.3%PinchBenchKilo Code10Pokee-Isaac 28B95.7%QwenClawBench10Qwen3.7 Max64.3%
Not in index
Artificial Analysis EnterpriseOps-GymArtificial Analysis23—ResearchClawBench19—Hermes Bench: 150 agent tasks graded on the files and state the agent leaves behindNous Research16—Artificial Analysis Harvey LAB-AA v1.012—Data Research and Analysis with Complex Operations10—Berkeley Function Calling Leaderboard v3Shishir G. Patil et al.9—Harvey LAB-AA v1.18—DeepPlanningDeepPlanning authors7—AA-Briefcase v1.1, rubric pass rate5—LifeVal: everyday life tasks5—MCP-Tasks5—RoboDojo SimAnthropic (reported run)5—WebArena-Verified Browser Agent BenchmarkAmine El Hattami et al.5—τ²-Bench Airline DomainVictor Barres et al.5—b3: Backbone Breaker BenchmarkLakera with the UK AI Security Institute4—CoWorkBench4—Finch (FinWorkBench): Finance & Accounting across Spreadsheet-Centric Enterprise WorkflowsFinch authors (ACL 2026 Findings)4—MLE-Bench Lite4—Refusal of harmful cyber requests (JailbreakBench, StrongREJECT, AgentHarm)Mistral AI (composite of JailbreakBench, StrongREJECT and AgentHarm)4—Remote Labor Index (RLI)Center for AI Safety; Scale AI4—SpreadsheetBench 24—Toolathlon Verified average assistant turns4—Toolathlon Verified Pass cubed4—Toolathlon Verified Pass@34—Agents' Last Exam, Linux subsetMeta (reported run)3—Anthropic CVE Rediscovery3—Artificial Analysis AutomationBench, strict pass@1Artificial Analysis3—BankerToolBench3—Legal Agent Benchmark all-pass rate — Harvey held-out set3—Legal Agent Benchmark mean criterion-pass rate — Harvey held-out set3—ToolSandboxApple3—WANDR (Wide ANd Deep Research)3—Agents' Last Exam v1.1, CLI subset2—CybenchAndy K. Zhang et al.2—GDPval rubrics2—General AI Assistants2—MCP-Atlas mean claim coverage2—MM-ClawBench2—OSWorld2—RelyToolBench2—terminalBench32—WebArena Web Agent BenchmarkShuyan Zhou et al.2—WebVoyager2—When2Call Multi-Turn (NVIDIA)2—AutomationBench, Zapier private evaluation set release 1.0.6Daniel Shepard and Robin Salimans1—BioMysteryBench (overall, v11 release)1—BrowseComp-PlusZijian Chen et al.1—Business Operation (Internal)1—CTI-REALM1—CWE-BenchCollinear1—DECK-Bench (Internal)1—EnterpriseOps-Gym (provider runs)ServiceNow1—Financial Modeling (Internal)1—Financial Research (Internal)1—Frontier Finance1—HANDBOOK.md1—Harvey's Legal Agent Benchmark, public 1,251-task set, all-pass1—Kimi Claw 24/7 Bench1—Kimi Hallucination Benchmark Hard with tools (Internal)1—Legal Agent Benchmark all-pass rate — Anthropic harness1—Legal Agent Benchmark mean criterion-pass rate — Anthropic harness1—LiveClawLiquid AI (reporting)1—LOCA-Bench 256KWeihao Zeng et al.1—MCPMark-VerifiedMCPMark1—MobileWorld1—Multi-Agent BrowseComp — 10-agent team prerelease configuration1—SWE Refactor Bench: Can Coding Agents Complete a Long-Horizon, Whole-Repository Stack Migration?Deyao Hong et al.1—Terminal-Bench-Science 0.1 with sixfold verifier timeout1—Coding
69 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
LiveCodeBenchVals AI141Claude Fable 5.190.5%Vibe Code Bench v1.1Vals AI123Claude Fable 5.293.8%Artificial Analysis SciCodeArtificial Analysis115Claude Fable 5.267.7%Software Engineering Benchmark VerifiedCarlos E. Jimenez et al.114Claude Opus 596.0%SWE-benchVals AI87Claude Opus 597.0%SWE-bench ProXiang Deng et al.83Claude Fable 5.293.0%Code MigrationVals AI77Claude Sonnet 5.569.8%LiveBench CodingLiveBench66Claude Sonnet 5.591.4%SWE-bench Verified · SWE-bench/experimentsSWE-bench team65Claude Opus 4.576.8%IOI v1Vals AI60Claude Opus 591.7%LiveCodeBench v6LiveCodeBench maintainers59Sakana Fugu-Ultra93.2%LiveBench Coding 2.0LiveBench52Claude Fable 598.3%SWE-bench Verified · swebench.comSWE-bench team46Claude 4.5 Opus74.4%IOIVals AI45Gemini 4 Argon100.0%MirrorCodeEpoch AI39Claude Sonnet 5.597.8%NL2Repo36DeepSeek V4.1 Flash 101768.2%Scientific Code Benchmark35Sakana Fugu60.1%SkillsBenchVals AI35DeepSeek V4.1 Flash 101769.8%SWE-Bench verifiedEpoch AI32Claude Opus 4.783.5%FrontierCode 2.027Claude Fable 5.249.1%Vibe Code Bench 1-100Vals AI26Muse Wonder Full34.0%Bug Hunt BenchPawel Huryn25Claude Fable 5.258.0%Evaluating Large Language Models Trained on CodeMark Chen et al.22o4-mini97.3%FrontierSWE v2Proximal22Claude Fable 5.269.9%cursorBench4021Gemini 4 Carbon60.4%ProgramBench: Can Language Models Rebuild Programs From Scratch?John Yang et al.19Claude Opus 593.0%cursorBench3218Claude Fable 5.173.4%FrontierCode 1.1 Main17Claude Opus 5.554.4%React Native EvalsCallstack16Composer 296.1%PostTrainBench v1.1PostTrainBench team15Claude Opus 5.549.3%SWE-bench Lite · swebench.comSWE-bench team15Claude 4 Sonnet56.7%VulcanBench v3VulcanBench contributors14Grok 4.589.9%OpenHarmony Bench v1.0OpenHarmony Bench authors13GLM-5.360.8%SWE-bench Lite · SWE-bench/experimentsSWE-bench team13Claude 4 Sonnet57.5%Artificial Analysis LiveCodeBenchArtificial Analysis12Gemini 3 Pro91.7%LiveCodeBench: Holistic and Contamination Free Evaluation of Large Language Models for CodeNaman Jain et al.12Solar Pro 487.8%ScarfBench: Cross-Framework Application Migration in Enterprise JavaAdvait Pavuluri et al.11Gemini 4 Argon75.2%SWE-bench Verified (mini-swe-agent-v2)5Claude Opus 4.675.6%
Not in index
ProgramBenchVals AI64Claude Fable 5.219.0%SWE-bench Multilingual · SWE-bench/experimentsSWE-bench team14Gemini 3 Flash72.7%SWE-RebenchNebius13—LiveCodeBench ProLiveCodeBench Pro authors9—LiveCodeBench v5LiveCodeBench maintainers9—SWE-MarathonAbundant AI and BenchFlow9—FrontierCode 1.1 Extended8—SWE-Atlas Codebase QnA8—cursorBench317—SWE-bench (full test split)SWE-bench team7Claude Opus 4.552.6%MLS-Bench LiteMLS-Bench4—FrontierCode 2.0, Artificial Analysis evaluation3—FrontierSWEEvan Chu et al.3—Kimi Code Bench v23—PostTrain Bench3—PostTrainBench v1.2PostTrainBench authors3—VulcanBench Coding Intelligence Index v1VulcanBench contributors3—DeepSeek DSBench FullStack2—DeepSeek DSBench Hard2—LiveCodeBench Pass@1 with Chain-of-Thought2—PaperBench2—BigCodeBench1—EEBench V1 core corpusatopile1—KernelBench Hard1—Kimi Code Bench 3.0 Long-horizon (Internal)1—Multi-SWE Bench1—ProgramBench hidden-test pass rate after episode 1Yang et al.1—Spider 2.0-LiteSpider 2.0 authors1—SVG-Bench1—VIBE V21—VIBE-Pro1—Reasoning
37 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Artificial Analysis Long Context ReasoningArtificial Analysis218Kimi K388.7%Critical Physics TasksArtificial Analysis215Claude Fable 5.235.4%ARC-AGI-2 (semi-private)ARC Prize Foundation131GPT-6 Astra95.0%ARC-AGI-1 (semi-private)ARC Prize Foundation126Gemini 4 Argon98.5%jevBench15104classifier.dev (fast)74.7%Chess PuzzlesEpoch AI90Claude Fable 5.279.0%jevBench1489classifier.dev (fast)70.8%LiveBench Data AnalysisLiveBench80Gemini 4 Argon85.2%LiveBench ReasoningLiveBench80Gemini 4 Argon95.1%Mystery Game PuzzlesEpoch AI80Claude Fable 5.290.0%Furniture AssemblyEpoch AI37Claude Fable 5.290.0%EBR-benchEpoch AI31Claude Fable 5.290.5%ARC-AGI-3 (semi-private)ARC Prize Foundation26Claude Fable 5.265.7%LongBench v2LongBench v2 authors25Inkling 2 Small70.6%Medical Long Context Reasoning (MLCR-AA)Wisedocs and Artificial Analysis22Claude Fable 5.243.9%MRCRv214Sakana Fugu-Ultra93.6%
Not in index
ARC-AGI-2 (public eval)ARC Prize Foundation122Claude Fable 5.298.8%ARC-AGI-1 (public eval)ARC Prize Foundation116Gemini 4 Argon99.5%STEMBench120: six-domain scientific research progressUlam9—BIG-Bench HardMirac Suzgun et al.8—BioBasicsBench: everyday wet-lab scienceOpenAI, Ligo7—Discrete Reasoning Over Paragraphs6—Testing the Limits of Chain-of-thought with Multistep Soft ReasoningZayne Sprague et al.6—GraphWalks BFS 256K–1M5—AI-Needle4—MRCR 1M4—Anthropic STEM Professor Bench3—CorpusQA 1M2—GeneBench-Pro2—Graphwalks BFS 0K-128K2—LongMemEvalLongMemEval authors2—OpenAI MRCR v2 8-needle 128K-256K2—OpenAI MRCR v2 8-needle 256K-512K2—OpenAI MRCR v2 8-needle 512K-1M2—OpenAI MRCR v2 8-needle 64K-128K2—RULER at a 64K-token contextRULER authors2—RULER at 128K context1—Multimodal
70 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Artificial Analysis MMMU-ProArtificial Analysis114Claude Opus 5.587.7%MMMU ProVals AI92Claude Fable 5.190.6%Massive Multi-discipline Multimodal Understanding ProMMMU-Pro authors53GPT-6.1 Terra85.5%CharXiv Reasoning49Claude Mythos 593.5%ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Boyang Zhang et al.39Claude Opus 5.578.8%Massive Multi-discipline Multimodal UnderstandingMMMU authors28Gemini 3 Pro87.2%ExtractBench: schema-guided enterprise document extraction (LlamaIndex leaderboard)Boyang Zhang et al.26GPT-6.1 Sol95.3%MathVision24Qwen3.8 Max95.2%ScreenSpot ProKaixin Li et al.20GPT-6 Astra92.7%ERQA18Qwen3.8 Max77.8%RealWorldQA18Qwen3.8-Flash-Next88.5%SimpleVQA18Qwen3.7 Plus81.7%CharXiv Reasoning without tools17Kimi K3.1 Pro89.0%OfficeQA ProOfficeQA Pro authors16Claude Fable 5.272.9%V*14Kimi K2.696.9%BabyVisionMeta AI11Qwen3.8 Max82.0%CountBench10Qwen3.6-27B97.8%RefCOCO averageRefCOCO dataset authors10Qwen3.6-27B92.5%ZeroBench10Muse Spark Open50.0%
Not in index
VideoMMMU13—Video-MME with subtitle12—OmniDocBench 1.511—MLVU mean average10—MMMU-Pro with Python10—Chartography with image and code toolsSurge AI and Anthropic9—Chartography without toolsSurge AI and Anthropic9—AI2D test split8—MedXpertQA Multimodal8—Multimodal Multi-disciplinary Video UnderstandingMMVU benchmark maintainers8—SWE-bench Multimodal · swebench.comSWE-bench team8o336.0%Video-MME without subtitle8—BabyVision with Python7—CC-OCR7—DynaMath7—ExtractBench Pro: long-tail document extraction challenges (LlamaIndex)LlamaIndex7—LVBench7—SWE-bench Multimodal · SWE-bench/experimentsSWE-bench team7o336.0%BenchCAD Vision2Code voxel IoU with toolsZhang et al. and Anthropic6—MathVision with Python6—MMLongBench-Doc6—OCRBench V2OCRBench authors6—We-Math6—BenchCAD Vision2Code voxel IoU without toolsZhang et al. and Anthropic5—OfficeQA5—PerceptionBench (Internal)5—Vision2Web5—ZeroBench_main with Python5—Anthropic biomedical-image-analysis evaluation4—Blueprint-Bench 2Google DeepMind4—ODINW134—Video-MMEVideo-MME benchmark team4—BenchCAD 2.1 Core, with tools3—BenchCAD 2.1 Core, without tools3—GDP.pdf mean criteria pass rate without tools3—PerceptionBench with toolsMeta (reported run)3—GDP.pdf mean criteria pass rate with tools2—Liquid image-to-JSON extraction JSON validity2—Liquid image-to-JSON extraction schema consistency F12—Liquid image-to-JSON extraction VLM judge score2—OfficeQA Pro v22—A Benchmark for Visual Question Answering using World KnowledgeDustin Schwenk et al.1—CharXiv Descriptive and Reasoning Combined1—GDP.pdf strict pass rate, page images plus text, Gemini judge1—GDP.pdf task-macro rubric pass rate with Python, Gemini judge1—MMSearch-Plus1—MStar1—olmOCR-BenchAllen Institute for AI1—OmniDocBench1—OmniDocBench v1.6Linke Ouyang et al.1—WorldVQA ForceAnswer1—Knowledge
58 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Artificial Analysis Humanity's Last ExamArtificial Analysis223Claude Fable 5.274.3%GPQA diamondEpoch AI217Gemini 4 Carbon96.5%Artificial Analysis GPQA DiamondArtificial Analysis200GPT-6 Astra96.1%Artificial Analysis Omniscience AccuracyArtificial Analysis165Claude Fable 5.279.1%Graduate-Level Google-Proof Q&A141GPT-6 Astra96.0%GPQA DiamondVals AI134Gemini 3.1 Pro Preview95.5%MMLU ProVals AI134Claude Fable 5.192.4%SimpleQA VerifiedEpoch AI91Claude Fable 5.294.0%GPQA Diamond · BenchLM84GPT-6 Astra96.0%Humanity's Last ExamCenter for AI Safety et al.77Claude Fable 5.274.2%Massive Multitask Language Understanding ProfessionalYubo Wang et al.75Qwen3.7 Max89.6%LiveBench LanguageLiveBench66Claude Fable 590.7%Humanity's Last Exam without tools53Claude Fable 5.273.2%Massive Multitask Language UnderstandingDan Hendrycks et al.34o191.8%SuperGPQA: Scaling LLM Evaluation Across 285 Graduate DisciplinesXiaoxuan Du et al.22Claude Opus 4.695.0%MMLU-Redux19Claude Opus 4.596.6%HealthBench ProfessionalRebecca Soskin Hicks et al.18Claude Sonnet 5.569.2%Artificial Analysis MMLU-ProArtificial Analysis15Gemini 3 Pro89.8%MMMLU15Interfaze Beta90.9%Measuring Short-Form Factuality in Large Language ModelsJason Wei et al.14GPT-4.562.5%HealthBench Hard13Muse Spark42.8%C-EvalC-Eval authors12Kimi K2.594.0%MMLU-Pro first-party comparison snapshot6Claude Opus 4.689.1%
Not in index
HealthBench length-adjusted score8—HealthBench Professional raw score8—HealthBench raw score8—Artificial Analysis Openness IndexArtificial Analysis7—LABBench2: An Improved Benchmark for AI Systems Performing Biology ResearchJon M. Laurent et al.7—Artificial Analysis HealthBench length-adjusted scoreOpenAI (benchmark); Artificial Analysis (evaluation)6—Artificial Analysis HealthBench Professional length-adjusted scoreOpenAI (benchmark); Artificial Analysis (evaluation)6—Artificial Analysis HealthBench Professional raw scoreOpenAI (benchmark); Artificial Analysis (evaluation)6—Artificial Analysis HealthBench raw scoreOpenAI (benchmark); Artificial Analysis (evaluation)6—BioMysteryBench Human Difficult6—HLE-VerifiedWeiqi Zhai et al.6—MedXpertQA Text6—Axiom Bio morphology-to-molecule matching5—BioMysteryBench Human Solvable5—FrontierScience Research5—LatchBio SingleCellBench5—Molecular Biology Protocols Troubleshooting5—PhysicianBench5—Anthropic de novo protein-binder design evaluation4—Anthropic medicinal-chemistry evaluation4—Anthropic Protein Design evaluation4—Benchling Molecular Biology Protocols Understanding3—LatchBio SpatialBench Verified3—Protein domain identification3—Protein mutation attribution3—Anthropic protein-design library-ranking task2—Chinese-SimpleQA2—FrontierScience2—GaRAGeAmazon Science2—RAGTruth: provider-reported F1RAGTruth authors2—AGIEval1—Anthropic Organic Chemistry V2 evaluation1—Chinese Massive Multitask Language Understanding1—Kimi Hallucination Benchmark without tools (Internal)1—ProteinGym Hard1—Multilingual
23 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
EuroEval PolishEuroEval200Gemini 3.7 Flash67.9%EuroEval SwedishEuroEval187GPT-6 Astra77.6%EuroEval GermanEuroEval183GPT-6 Astra68.0%EuroEval PortugueseEuroEval174GPT-6.1 Sol74.8%EuroEval FrenchEuroEval170Gemini 3.7 Flash77.1%EuroEval DutchEuroEval168GPT-6 Astra74.7%EuroEval SpanishEuroEval168GPT-6 Astra70.0%EuroEval ItalianEuroEval166GPT-6.1 Sol78.5%MMLU-ProXMMLU-ProX authors17Qwen3.7 Max87.0%Multilingual Grade School MathFreda Shi et al.13o4-mini93.7%INCLUDE12Claude Opus 589.8%
Not in index
MGSMVals AI73Claude Opus 4.5 Thinking95.2%Artificial Analysis Global-MMLU-LiteArtificial Analysis14—SWE-bench Multilingual · swebench.comSWE-bench team12Gemini 3 Flash72.7%NOVA-637—PolyMath6—Global MMLUSingh et al.5—MAXIFE5—Multi-task Indic Language Understanding BenchmarkVerma et al.5—Long-context translation evaluation (Cohere)1—MKQA-11 multilingual retrieval1—NanoBEIR Multilingual Extended1—WMT26 General Machine Translation (all languages)WMT organizers1—Instruction
17 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Artificial Analysis IFBenchArtificial Analysis147MiniMax M382.9%Instruction-Following EvalJeffrey Zhou et al.70LFM2.5-24B-A2B96.8%LiveBench Instruction FollowingLiveBench66Gemini 3.8 Flash81.4%Instruction Following Benchmark57Flint89.2%
Not in index
MultiChallenge: provider-reported runs3—AgentIF: instruction following in agentic scenariosAgentIF authors2—Brevity Benchmark (NVIDIA)2—EQ-Bench 3EQ-Bench2—IHEval: instruction hierarchy evaluationIHEval authors2—InverseIFEvalInverseIFEval authors2—RoleBench (RoleLLM)RoleLLM authors2—RoleMRCRoleMRC authors2—StructEval text-format subsetStructEval authors2—SysBench: system message followingSysBench authors2—IFStruct1—Multi-IF1—Structured Output Benchmark Value Accuracy1—Math
39 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
OTIS Mock AIME 2024-2025Epoch AI200Gemini 4 Carbon100.0%AIMEVals AI94Gemini 3.1 Pro Preview98.1%FrontierMath-Tiers-1-3-v2-PrivateEpoch AI90Claude Fable 5.295.1%MATH level 5Epoch AI88GPT-598.1%FrontierMath-Tier-4-v2-PrivateEpoch AI72Claude Fable 5.2100.0%FrontierMath-2025-02-28-PrivateEpoch AI69GPT-5.551.7%LiveBench MathematicsLiveBench66Claude Opus 5.597.1%MATH 500Vals AI59Gemini 3 Pro96.4%FrontierMath-Tier-4-2025-07-01-PrivateEpoch AI55AI Co-Mathematician47.9%ProofBench v1.1Vals AI54Claude Fable 5.2100.0%American Invitational Mathematics Examination 202553GPT-5.2100.0%AIME 202634GLM-5.299.2%American Invitational Mathematics Examination 202427Ripple-65B Preview97.1%Harvard-MIT Mathematics Tournament February 202627Qwen3.7 Max97.1%MATH-500 Problem SetDan Hendrycks et al.27Ternary Bonsai 2 27B98.8%Harvard-MIT Mathematics Tournament February 202515GLM-597.5%Harvard-MIT Mathematics Tournament November 202512GLM-596.9%IMOAnswerBench10dots3-note Preview90.9%Artificial Analysis MATH-500Artificial Analysis9GPT-5 (high)99.4%
Not in index
FrontierMath-2025-02-28-PublicEpoch AI64Claude Fable 5100.0%FrontierMath-Tier-4-2025-07-01-PublicEpoch AI36Claude Fable 5100.0%FrontierMath-ErdosEpoch AI15GPT-6.1 Sol4.4%Artificial Analysis AIME 2025Artificial Analysis12—MMAnswerBench10—Apex9—Grade School Math 8KKarl Cobbe and colleagues (OpenAI)7—AIME25 first-party comparison snapshot6—Apex Shortlist5—OEIS Open LiteEpoch AI5Claude Fable 544.0%OEIS OpenEpoch AI3Claude Opus 4.829.9%United States of America Mathematical Olympiad 2026Mathematical Association of America3—ArXivMath August 2026 with tools2—ArXivMath August 2026 without tools2—International Mathematical Olympiad 20262—ArXivMath June 2026 with tools1—ArXivMath June 2026 without tools1—International Physics Olympiad 2025 (Theory)1—RiemannBench with tools1—RiemannBench without tools1—Legal
4 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Finance
7 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Healthcare
3 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Education
1 benchmarkBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Not in index
SAGEVals AI29Claude Opus 5.591.3%Science
3 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Public benefits
2 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Other
8 benchmarksBenchmarkModelsModels with a result.Top resultBest result on this benchmark.
Not in index
CyberBench v1.1Vals AI48Muse Wonder80.4%SRE BenchVals AI36GPT-6 Astra56.9%Time Horizon Index: KSPVals AI13Claude Opus 5.591.3%Vals Web Search IndexVals AI12Claude Opus 5.5 Parallel Fast57.5%CUA-benchVals AI9GPT-6 Astra19.2%SpireClimbEpoch AI9GPT-6 Astra34.0%SpireClimb (Slay the Spire 1, Ascension 20 Heart)Epoch AI9GPT-6 Astra33.3%SpireClimb (Slay the Spire 2, Ascension 10)Epoch AI9GPT-6 Astra34.7%