- Agentic scientific research Terminal-Bench-Science 0.1 52.6
- Knowledge work GDPval-AA v2 1853
- Computer use OSWorld 2.0 (partial) 77.9
- Multidisciplinary reasoning Humanity's Last Exam (with tools) 65
- Business workflows AutomationBench 31.4
- Agentic coding CursorBench 3.2.0 73.4
- SWE-bench Pro 81.2
- SWE-bench Multilingual 89.1
- SWE-bench Multimodal 54.7
- HealthBench Professional 62.1
- AA-Briefcase 1694
- ARC-AGI-1 97.5
- ARC-AGI-2 90
Claude Fable 5.1
Claude Fable 5.1 ranks first on 31 of the 334 benchmarks with five or more models scored.
342
Benchmarks with a score
31
Ranked first, out of five or more models
113
In the top three
Anthropic Sep 2026
Claude Fable 5.1
Closed weights Flagship
167.0
#7 of 256
Epoch Capabilities Index
Self-Reported Index 167.9 #8 of 404
- Humanity's Last Exam 46.5% #2 /47
- AutomationBench 22.4% #15 /42
- Terminal-Bench 4.0 57.9% #6 /40
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Longform Creative Writing (EQ-Bench) | 85.3 | #2 of 119 | Official leaderboard | |
| Lech Mazur Creative Writing | 3.75 | #1 of 50 | Official leaderboard | |
| LMArena (Elo) | 1,503 | #7 of 335 | Official leaderboard | |
| Creative Writing v3 | 81.3 | #5 of 120 | Official leaderboard | |
| Humanity's Last Exam | 46.5% | #2 of 47 | Official leaderboard | |
| BullshitBench V2 | 86.5% | #6 of 129 | Official leaderboard | |
| Document Arena | 1,498 | #2 of 39 | Official leaderboard | |
| Hemingway-bench | 1,095 | #3 of 42 | Official leaderboard | |
| DesignArena ASCII Art | 1,318 | #6 of 77 | Official leaderboard | |
| BullshitBench | 71.7% | #9 of 107 | Official leaderboard | |
| SimpleQA Verified | 70.8% | #8 of 94 | Epoch AI run | |
| HANDBOOK.md | 47.1% | #3 of 16 | Official leaderboard | |
| Last Translation Benchmark | 32.9 | #7 of 36 | Official leaderboard | |
| AA-LCR Extended | 73% | #10 of 49 | Official leaderboard | |
| Global MMLU | 94% | #3 of 14 | Lab model card | |
| StudyArena Text | 1,263 | #19 of 85 | Official leaderboard | |
| Extended NYT Connections | 90% | #24 of 103 | Official leaderboard | |
| Joke Arena | 1,545 | #6 of 25 | Official leaderboard | |
| Antidote | 1,065 | #9 of 36 | Official leaderboard | |
| MILU | 93% | #3 of 11 | Lab model card | |
| HLE-Diamond | 55.1% | #7 of 20 | Official leaderboard | |
| ELI5 Arena | 1,567 | #7 of 18 | Official leaderboard | |
| Corp Speak Arena | 1,541 | #10 of 22 | Official leaderboard | |
| ComplexConstraints | 60.4% | #9 of 15 | Official leaderboard | |
| Kimi Hallu Bench | 11.4% | #4 of 6 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ProofBench v1.1 | 100% | #1 of 52 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 100% | #5 of 212 | Epoch AI run | |
| DTBench | 97.6% | #5 of 180 | Official leaderboard | |
| SimpleBench | 86.6% | #5 of 106 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 90.2% | #7 of 86 | Official leaderboard | |
| ARC-AGI-2 | 90% | #9 of 98 | Official leaderboard | |
| BixBench3 | 51.5 | #2 of 20 | Official leaderboard | |
| BioMysteryBench | 90.3% | #1 of 10 | Lab model card | |
| LatchBio Capabilities v1.0 | 53% | #3 of 27 | Official leaderboard | |
| ArXivMath | 91.3% | #1 of 9 | Lab model card | |
| ARC-AGI | 97.5% | #12 of 95 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 87.8% | #9 of 68 | Official leaderboard | |
| EurekaBench | 29.8 | #1 of 7 | Official leaderboard | |
| SpatialBench (LatchBio) | 77.6% | #1 of 7 | Lab model card | |
| ForecastBench | 0.12 | #13 of 82 | Official leaderboard | |
| MysteryMechanism | 47.8% | #5 of 31 | Official leaderboard | |
| SciUniverse | 45.3% | #1 of 6 | Official leaderboard | |
| Terminal-Bench-Science | 40% | #3 of 17 | Official leaderboard | |
| scBench (LatchBio SingleCellBench) | 60.5% | #5 of 26 | Official leaderboard | |
| PaperBenchX | 12.9% | #2 of 10 | Official leaderboard | |
| Prophet Arena | 86.0 | #5 of 24 | Official leaderboard | |
| FrontierPhysics | 11.3% | #6 of 22 | Official leaderboard | |
| ScholarCatalyst | 53% | #1 of 3 | Official leaderboard | |
| Leipzig Benchmark | 79% | #7 of 19 | Official leaderboard | |
| Riemann-bench | 71.2% | #8 of 20 | Official leaderboard | |
| Organic Chemistry V2 (Anthropic) | 69.2% | #1 of 2 | Lab model card | |
| CritPt | 72% | #7 of 11 | Official leaderboard | |
| CritPt v2 | 72% | #7 of 11 | Official leaderboard | |
| FrontierMath Erdős | 0% | #13 of 15 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| WeirdML | 92.9% | #3 of 112 | Official leaderboard | |
| GSO-Bench | 88.2% | #1 of 30 | Official leaderboard | |
| DesignArena Data Visualization | 1,346 | #4 of 118 | Official leaderboard | |
| SWE Atlas Test Writing | 67% | #1 of 22 | Official leaderboard | |
| SWE-Together | 69.3% | #1 of 22 | Official leaderboard | |
| ALE-Bench | 2,849 | #6 of 126 | Official leaderboard | |
| WebDev Arena | 1,758 | #5 of 105 | arena.ai, via Epoch AI | |
| PostTrainBench 1.2 | 44.6% | #1 of 21 | Official leaderboard | |
| SWE-Bench Pro | 81.2% | #4 of 81 | Lab model card | |
| Image-to-WebDev Arena | 1,724 | #3 of 55 | Official leaderboard | |
| Code Arena | 1,742 | #7 of 128 | Official leaderboard | |
| DesignArena Frontend Web App | 1,315 | #3 of 53 | Official leaderboard | |
| Vibe Code Bench | 90.3% | #7 of 120 | Official leaderboard | |
| APEX-SWE | 63.6% | #4 of 65 | Official leaderboard | |
| MLS-Bench | 50.3% | #1 of 15 | Official leaderboard | |
| DesignArena Agentic Overall | 1,327 | #5 of 74 | Official leaderboard | |
| SWE-bench Multilingual | 89.1% | #5 of 73 | Lab model card | |
| DesignArena Game Dev | 1,365 | #10 of 143 | Official leaderboard | |
| ReactBench | 49.2% | #2 of 28 | Official leaderboard | |
| Next.js Agent Evals | 100% | #1 of 14 | Official leaderboard | |
| DesignArena 3D Design | 1,391 | #10 of 136 | Official leaderboard | |
| MirrorCode | 93.3% | #3 of 39 | Epoch AI run | |
| ProgramBench | 9.00% | #2 of 25 | Official leaderboard | |
| DesignArena Full Stack | 1,304 | #4 of 49 | Official leaderboard | |
| DesignArena Godot Game Dev | 1,307 | #2 of 24 | Official leaderboard | |
| DesignArena Code | 1,331 | #13 of 143 | Official leaderboard | |
| SWE Atlas | 60% | #2 of 22 | Official leaderboard | |
| TasteVal | 171.6× | #2 of 20 | Official leaderboard | |
| SWE Atlas Codebase QnA Extended | 38% | #4 of 38 | Official leaderboard | |
| SWE Atlas v1.1 | 76.4% | #3 of 28 | Official leaderboard | |
| DesignArena UI Component | 1,333 | #15 of 139 | Official leaderboard | |
| Vals RSI Index | 0.36 | #3 of 27 | Official leaderboard | |
| SWE-Bench Pro V2 Hard | 92.2% | #2 of 18 | Official leaderboard | |
| DesignArena Website | 1,306 | #17 of 151 | Official leaderboard | |
| DesignArena SVG | 1,294 | #12 of 103 | Official leaderboard | |
| SWE Atlas Refactoring | 56.7% | #2 of 17 | Official leaderboard | |
| ProgramBench almost solved | 55.5% | #3 of 25 | Official leaderboard | |
| DesignArena Image to Frontend | 1,260 | #4 of 31 | Official leaderboard | |
| IOI v2 (Vals) | 90.8% | #6 of 45 | Official leaderboard | |
| Code Migration | 54.6% | #11 of 78 | Official leaderboard | |
| DesignArena React Native | 1,256 | #8 of 56 | Official leaderboard | |
| DAXBench v2 | 90.7% | #7 of 47 | Official leaderboard | |
| Vibe Code Bench 1-100 | 28% | #4 of 26 | Official leaderboard | |
| SWE-Bench Pro V2 | 99.1% | #3 of 18 | Official leaderboard | |
| AutoResearchExam | 0.60 | #2 of 12 | Official leaderboard | |
| TestPrism | 28% | #1 of 6 | Official leaderboard | |
| SWE-bench CLI | 14.5% | #3 of 16 | Official leaderboard | |
| ProgramBench tests passed | 82.5% | #5 of 25 | Official leaderboard | |
| DAYJOB: Data Science | 26% | #3 of 15 | Official leaderboard | |
| WeirdML v3 | 26% | #6 of 27 | Official leaderboard | |
| RSI-Exam | 48.1% | #4 of 18 | Official leaderboard | |
| MiMo Visual Coding | 74.4% | #2 of 9 | Official leaderboard | |
| FrontierSWE | 56.3% | #6 of 25 | Official leaderboard | |
| DeepRSI | 1,548 | #2 of 8 | Official leaderboard | |
| LibraryDesignBench | 47.5 | #2 of 8 | Official leaderboard | |
| Senior SWE-Bench | 34.7% | #6 of 23 | Official leaderboard | |
| Chart Arena | 1,598 | #4 of 15 | Official leaderboard | |
| Floorplan Arena | 1,713 | #3 of 11 | Official leaderboard | |
| sudo L7 | 32.2% | #5 of 18 | Official leaderboard | |
| FrontierCode 2.0 | 34% | #9 of 32 | Official leaderboard | |
| CursorBench | 51.8% | #6 of 21 | cursor.com, via Epoch AI | |
| SlopCodeBench | 27.6% | #7 of 22 | Official leaderboard | |
| CursorBench 4.0 | 51.8% | #7 of 21 | Official leaderboard | |
| Real-SWE 1.1 | 45.6% | #5 of 15 | Official leaderboard | |
| SWE-Clar | 60% | #3 of 9 | Official leaderboard | |
| SWE-bench Verified Extended | 61.3% | #16 of 44 | Official leaderboard | |
| Snorkel Agentic Coding 2.0 | 39.6% | #6 of 16 | Official leaderboard | |
| SWE-sweep | 2.90% | #4 of 10 | Official leaderboard | |
| Web Clone Bench | 13.3% | #4 of 10 | Official leaderboard | |
| SWE-bench Multimodal | 54.7% | #5 of 12 | Lab model card | |
| FrontierCS 2 | 35.4% | #8 of 19 | Official leaderboard | |
| Landing Page Arena | 1,556 | #6 of 14 | Official leaderboard | |
| DesignArena Image to HTML | 1,213 | #22 of 51 | Official leaderboard | |
| FrontierCode 1.1 | 50.9% | #6 of 13 | Lab model card | |
| DeepSWE | 67.4% | #26 of 56 | Lab model card | |
| Android Bench 2.0 | 22.7% | #8 of 17 | Official leaderboard | |
| SWE-Marathon v1.1 | 45.6% | #10 of 20 | Official leaderboard | |
| OWE-Bench | 29.1% | #7 of 14 | Official leaderboard | |
| Kimi Code Bench 3.0 Long-horizon | 58.8% | #3 of 6 | Official leaderboard | |
| CursorBench 3.2 | 73.4% | #1 of 2 | Lab model card | |
| OpenDesign Arena | 80.3 | #14 of 21 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LisanBench | 15100.3 | #6 of 133 | Official leaderboard | |
| OSWorld 2.0 | 77.9% | #1 of 19 | Lab model card | |
| ApprenticeBench | 72% | #1 of 18 | Official leaderboard | |
| DuelLab GameBench 2 | 94.0 | #4 of 57 | Official leaderboard | |
| MCP Atlas | 87.2% | #3 of 37 | Official leaderboard | |
| Agent Arena | 14.8% | #5 of 56 | Official leaderboard | |
| OSWorld-Science | 73.7% | #1 of 11 | Official leaderboard | |
| BrowseComp Extended | 63.3% | #4 of 42 | Official leaderboard | |
| Game Arena Nine Men's Morris | 1,001 | #4 of 39 | Official leaderboard | |
| Game Arena Go | 1,601 | #5 of 39 | Official leaderboard | |
| Game Arena Poker | 17.9 | #4 of 31 | Official leaderboard | |
| Mystery Game Puzzles | 58% | #11 of 84 | Epoch AI run | |
| RuneBench | 6.14 | #10 of 76 | Official leaderboard | |
| Toolathlon-Verified | 77.8% | #3 of 21 | Lab model card | |
| Game Arena Dots and Boxes | 1,096 | #6 of 40 | Official leaderboard | |
| Terminal-Bench 4.0 | 57.9% | #6 of 40 | Official leaderboard | |
| Game Arena Four in a Row | 608 | #7 of 43 | Official leaderboard | |
| Terminal-Bench 4.1 | 57.6% | #7 of 43 | Official leaderboard | |
| DRACO | 87.7% | #2 of 12 | Lab model card | |
| Game Arena Clobber | 519 | #7 of 39 | Official leaderboard | |
| Game Arena Lines of Action | 655 | #7 of 39 | Official leaderboard | |
| BabaIsBench | 100% | #6 of 32 | Official leaderboard | |
| LMArena Search Arena | 1,227 | #7 of 37 | Official leaderboard | |
| Kaggle Game Arena | 417 | #8 of 42 | Official leaderboard | |
| EBR-bench | 57.1% | #7 of 31 | Epoch AI run | |
| Game Arena Bargaining | 224 | #9 of 39 | Official leaderboard | |
| Time Horizon Index: KSP | 63.3% | #3 of 13 | Official leaderboard | |
| Chess Puzzles | 48% | #22 of 93 | Epoch AI run | |
| Game Arena Checkers | 1,086 | #10 of 41 | Official leaderboard | |
| Game Arena Word Art | 1,196 | #10 of 39 | Official leaderboard | |
| Clocktower Bench | 1,595 | #3 of 11 | Official leaderboard | |
| Game Arena Reversi | 743 | #11 of 39 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 439 | #11 of 39 | Official leaderboard | |
| Collaborative Intelligence Benchmark | 65.3% | #6 of 18 | Official leaderboard | |
| CUA-bench | 13.2% | #3 of 9 | Official leaderboard | |
| AutomationBench | 22.4% | #15 of 42 | Official leaderboard | |
| Realtime-Bench | 16.7% | #4 of 8 | Official leaderboard | |
| Game Arena Dark Hex | 252 | #22 of 41 | Official leaderboard | |
| Game Arena Chess | 761 | #23 of 42 | Official leaderboard | |
| StarSkirmish Bench | 37.2% | #5 of 9 | Official leaderboard | |
| Game Arena Coin Game | 330 | #22 of 39 | Official leaderboard | |
| BU Bench V2.1 | 72.0 | #10 of 16 | Official leaderboard | |
| SpireClimb | 2.00% | #7 of 9 | Epoch AI run | |
| SpireClimb StS1 A20 Heart | 0% | #7 of 9 | Official leaderboard | |
| SpireClimb StS2 A10 | 4.00% | #7 of 9 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 19.5% | #37 of 44 | Official leaderboard | |
| FocusBench (Meta) | 36.6% | #5 of 5 | Official leaderboard | |
| OSWorld 2.0 (Aug 2026) | 77.9% | #1 of 1 | Lab model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Excel Modeling Benchmark | 76.7% | #1 of 76 | Official leaderboard | |
| LegalBench | 88.5% | #2 of 138 | Official leaderboard | |
| EnterpriseBench: CoreCraft | 77.4% | #1 of 51 | Official leaderboard | |
| MedScribe | 91.3% | #2 of 97 | Official leaderboard | |
| MortgageTax | 70.8% | #2 of 87 | Official leaderboard | |
| DesignArena Spreadsheet | 1,400 | #1 of 35 | Official leaderboard | |
| Legal Research Bench | 55.3% | #3 of 93 | Official leaderboard | |
| Tax Agent Bench | 50.8% | #3 of 84 | Official leaderboard | |
| Public Benefits Bench | 74.9% | #2 of 50 | Official leaderboard | |
| Immigration Bench | 75.4% | #1 of 25 | Official leaderboard | |
| Vals Index v2 | 68.8% | #3 of 65 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 76.7% | #2 of 42 | Official leaderboard | |
| YC-Bench | $3,199,375 | #3 of 59 | Official leaderboard | |
| FrontierFinance | 55.9% | #1 of 17 | Official leaderboard | |
| Tuesday Work Index | 72.4 | #3 of 44 | Official leaderboard | |
| TaxEval v2 | 76% | #9 of 129 | Official leaderboard | |
| GDPval-AA v2 | 1,853 | #2 of 27 | Lab model card | |
| $OneMillion-Bench | 42.1% | #1 of 13 | Official leaderboard | |
| RealmPathology | 78% | #1 of 13 | Official leaderboard | |
| DAYJOB: Healthcare | 31.2% | #3 of 37 | Official leaderboard | |
| Vals Index v2.1 | 65.8% | #4 of 49 | Official leaderboard | |
| MedCode | 53.5% | #8 of 96 | Official leaderboard | |
| FoodTruck Bench | $62,732 | #5 of 59 | Official leaderboard | |
| MedXpertQA MM Extended | 66% | #4 of 41 | Official leaderboard | |
| Vals Finance Agent v2 | 58.9% | #8 of 80 | Official leaderboard | |
| Genspark Slides Benchmark | 83.6% | #1 of 10 | Official leaderboard | |
| DAYJOB: Finance | 36.5% | #4 of 39 | Official leaderboard | |
| DesignArena Python-PPTX Slides | 1,308 | #4 of 36 | Official leaderboard | |
| LMBuild | 65.9% | #2 of 18 | Official leaderboard | |
| APEX-1 | 73.8% | #8 of 71 | Official leaderboard | |
| APEX-Agents | 68.6% | #7 of 61 | Collected by Epoch AI, source not named | |
| APEX-Accounting | 66.7% | #6 of 49 | Official leaderboard | |
| DAYJOB: Tax | 9.60% | #2 of 15 | Official leaderboard | |
| DesignArena HTML Slides | 1,395 | #4 of 28 | Official leaderboard | |
| Holdings-IndividualTaxBench | 75.5% | #2 of 13 | Official leaderboard | |
| PolicyBench | 90.8% | #8 of 50 | Official leaderboard | |
| GDPval Extended | 3.80% | #7 of 43 | Official leaderboard | |
| Epoch Automation Reports | 65.3% | #1 of 6 | Official leaderboard | |
| Legora BAR | 104.0 | #3 of 17 | Official leaderboard | |
| ATLAS-Finance | 10.3% | #2 of 11 | Official leaderboard | |
| ATLAS-Finance v1.1.0 | 10.3% | #2 of 11 | Official leaderboard | |
| APEX-Agents 1.1 | 77.1% | #12 of 65 | Official leaderboard | |
| PRBench | 51.6% | #9 of 48 | Official leaderboard | |
| Digital Work Index | 23.6 | #3 of 16 | Official leaderboard | |
| GDP.xlsx | 23.1% | #4 of 20 | Official leaderboard | |
| Thrive Holdings helpdesk ticket resolution benchmark | 55.8% | #3 of 15 | Official leaderboard | |
| LegalReasonBench | 50.6% | #2 of 10 | Official leaderboard | |
| Build-A-Chip-Bench | 16.6% | #1 of 5 | Official leaderboard | |
| Remote Labor Index | 17.9% | #4 of 19 | Official leaderboard | |
| OfficeQA | 80.2% | #3 of 14 | Lab model card | |
| OfficeQA Pro | 69% | #4 of 16 | Lab model card | |
| SnorkelFinance 2.0 | 19.6% | #4 of 16 | Official leaderboard | |
| SnorkelRevOps | 14% | #4 of 16 | Official leaderboard | |
| Clipping Bench | $1,068 | #5 of 19 | Official leaderboard | |
| Parametric CAD Bench | 56.8% | #3 of 11 | Official leaderboard | |
| WorkplaceAgents | 15.8% | #4 of 14 | Official leaderboard | |
| AA-Briefcase | 1,694 | #2 of 7 | Lab model card | |
| DAYJOB: Accounting | 27.2% | #5 of 17 | Official leaderboard | |
| GDPx | 8.53% | #6 of 20 | Official leaderboard | |
| DeepFinance | 45.6% | #7 of 23 | Official leaderboard | |
| CAD Arena | 0.66 | #4 of 12 | Official leaderboard | |
| FrontierCAD Assemblies | 0.62 | #2 of 6 | Official leaderboard | |
| FrontierCAD Parts | 0.58 | #2 of 6 | Official leaderboard | |
| TaxCalcBench v2 | 58% | #12 of 32 | Official leaderboard | |
| DAYJOB: Law | 26.5% | #6 of 16 | Official leaderboard | |
| HealthBench | 60% | #12 of 31 | Lab model card | |
| MBABench 2.0 | 7.90% | #2 of 5 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 6.67% | #33 of 80 | Official leaderboard | |
| EEBench | 56.4% | #13 of 30 | Official leaderboard | |
| Vending-Bench Arena | $5,719 | #14 of 32 | Official leaderboard | |
| SnorkelLegal | 35.9% | #7 of 16 | Official leaderboard | |
| SnorkelManufacturing | 11.3% | #7 of 16 | Official leaderboard | |
| SnorkelUnderwrite 2.0 | 24% | #7 of 16 | Official leaderboard | |
| HealthBench Professional | 62.1% | #12 of 26 | Lab model card | |
| PharmaBench (micro1) | 12.8% | #6 of 13 | Official leaderboard | |
| Vending-Bench 2 | $5,422 | #31 of 67 | Official leaderboard | |
| MissionReport | 63.3% | #5 of 10 | Official leaderboard | |
| HW-Bench | 23% | #8 of 15 | Official leaderboard | |
| RWE-bench | 7.00% | #8 of 15 | Official leaderboard | |
| micro1 Medical Index | 68.2% | #5 of 8 | Official leaderboard | |
| LifeSci Diamond | 1.80% | #14 of 18 | Official leaderboard | |
| BenchCAD 2.0 | 30% | #4 of 5 | Official leaderboard | |
| BenchCAD 2.1 | 44.4% | #4 of 5 | Official leaderboard | |
| BenchCAD | 43.7% | #9 of 11 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ParseBench | 77.8% | #2 of 44 | Official leaderboard | |
| InteractiveBench | 41.7% | #1 of 12 | Official leaderboard | |
| ArtifactArena | 1,310 | #2 of 21 | Official leaderboard | |
| Vision Arena | 1,289 | #13 of 136 | Official leaderboard | |
| Chartography | 46.2% | #5 of 43 | Official leaderboard | |
| MineBench | 1,909 | #11 of 75 | Official leaderboard | |
| MMMU-Pro Extended | 47.6% | #6 of 38 | Official leaderboard | |
| Drone-Bench | 91.2% | #3 of 19 | Official leaderboard | |
| 4DCodeBench | 0.74 | #3 of 16 | Official leaderboard | |
| GeoBench | 84% | #10 of 48 | Official leaderboard | |
| Blueprint-Bench 2 | 41.9% | #8 of 38 | Official leaderboard | |
| ClockBench | 59.4% | #8 of 37 | Official leaderboard | |
| ScienceVision | 0.46 | #3 of 12 | Official leaderboard | |
| Furniture Assembly Benchmark | 70% | #9 of 35 | Official leaderboard | |
| Humanity's Sixth Sense | 40.8% | #9 of 34 | Official leaderboard | |
| GDP.pdf | 49.4% | #6 of 19 | Official leaderboard | |
| Roboharm | 20% | #1 of 3 | Official leaderboard | |
| CharXiv Extended | 82.2% | #14 of 38 | Official leaderboard | |
| Code2Video Benchmark | 1,536 | #8 of 19 | Official leaderboard | |
| Code2Video Arena | 1,527 | #6 of 13 | Official leaderboard | |
| GlobeBench | 21.6% | #10 of 19 | Official leaderboard | |
| Dual-Arm Sim | 37.8 | #6 of 11 | Official leaderboard | |
| Robo-bench | 5.83% | #5 of 7 | Official leaderboard | |
| Pantograph frontier-LLM robot benchmark (Pandroid) | 15% | #2 of 2 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| nocot-bench (NCRI) | 128.0 | #5 of 185 | Official leaderboard | |
| Single-turn benign request evaluations (Anthropic) | 0% | #1 of 19 | Lab model card | |
| OncologyBench | 44% | #2 of 30 | Official leaderboard | |
| EpiBench | 31.8% | #2 of 27 | Official leaderboard | |
| MetagenomicsBench | 61.7% | #2 of 26 | Official leaderboard | |
| TxBench-Oligo-Discovery | 54.9% | #3 of 30 | Official leaderboard | |
| BioSecBench-Function | 49% | #3 of 27 | Official leaderboard | |
| PCT (Polarized Contrastive Pairs) | 33.4% | #6 of 46 | Official leaderboard | |
| TxBench-Antibody-Discovery | 63.3% | #4 of 30 | Official leaderboard | |
| Prompt injection robustness against adaptive attackers, computer use (Anthropic) | 0.07% | #2 of 14 | Lab model card | |
| AI Wellbeing Index percent positive experiences | 77.3% | #1 of 6 | Official leaderboard | |
| FORTRESS | 13.7% | #13 of 73 | Official leaderboard | |
| Gray Swan IPI Arena benchmark | 0.10% | #2 of 11 | Lab model card | |
| BBQ (Bias Benchmark for QA) | 89.9% | #5 of 25 | Lab model card | |
| Cyber harm coverage (Anthropic) | 100% | #1 of 5 | Lab model card | |
| SRE-Bench | 22.9% | #9 of 35 | Official leaderboard | |
| Virology Capabilities Test (VCT) | 58% | #8 of 31 | Lab model card | |
| BioSecBench-Surveillance | 49% | #7 of 27 | Official leaderboard | |
| Anthropic political even-handedness eval | 98.3% | #4 of 14 | Lab model card | |
| Malicious use of Claude Code (Anthropic) | 90.3% | #5 of 17 | Lab model card | |
| Biopharma Bench | 57.4% | #4 of 13 | Official leaderboard | |
| KORA | 57.5% | #13 of 42 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 59.7% | #10 of 30 | Official leaderboard | |
| scBench-Long | 25.6% | #9 of 27 | Official leaderboard | |
| VariantBench | 40.7% | #9 of 27 | Official leaderboard | |
| BrowseSafe++ | 86.5% | #6 of 15 | Official leaderboard | |
| CheatBench | 45.8% | #4 of 10 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 70.4% | #20 of 44 | Official leaderboard | |
| SpatialBench-Long | 29.8% | #13 of 27 | Official leaderboard | |
| DistressBench | 71.5% | #21 of 42 | Official leaderboard | |
| CWE-bench | 58% | #6 of 12 | Official leaderboard | |
| Rewind-attacker cyber adversarial robustness (Anthropic) | 4.50% | #3 of 6 | Lab model card | |
| Malicious computer use (Anthropic) | 85.7% | #9 of 14 | Lab model card | |
| Prompt injection robustness against adaptive attackers, browser use (Anthropic) | 2.64% | #9 of 13 | Lab model card | |
| Suicide Memory Bench | 0.63 | #9 of 12 | Official leaderboard | |
| WellnessBench | 51% | #9 of 12 | Official leaderboard | |
| BioSecBench-Refusal | 5.60% | #17 of 22 | Official leaderboard | |
| Perceptual Attacks on Embodied AI (Robocurve) | 40% | #4 of 5 | Official leaderboard | |
| Prompt injection robustness against adaptive attackers, coding (Anthropic) | 9.26% | #12 of 14 | Lab model card | |
| Single-turn harmful request evaluations (Anthropic) | 94.7% | #17 of 19 | Lab model card | |
| Secure and defensive coding (Anthropic) | 1.03% | #2 of 2 | Lab model card | |
| Gray Swan Shade knowledge elicitation through chat | 1.00% | #1 of 1 | Lab model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Conceptual Reasoning Index | 74.7 | #4 of 151 | Official leaderboard | |
| LMCA | 65.5% | #4 of 147 | Official leaderboard | |
| ACCoRD | 83.7 | #6 of 152 | Official leaderboard | |
| CAIS Risk Index | 30.3 | #4 of 67 | Official leaderboard | |
| LiveBench | 84% | #3 of 31 | Official leaderboard | |
| CAIS Text Capabilities Index | 55.6 | #7 of 68 | Official leaderboard | |
| CAIS Vision Capabilities Index | 68.9 | #7 of 53 | Official leaderboard | |
| SAGE (Vals) | 48.5% | #27 of 81 | Official leaderboard | |
| Xbench | 24.9 | #12 of 17 | Official leaderboard |
Its lab's numbers put it at 167.9, in line with its ECI
The Self-Reported Index fits one score to the 28 capability results Anthropic reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 164.3 to 171.4. How we compute it.
- ARC-AGI 97.5
- Global MMLU 94
- MILU 93
- ArXivMath 91.3
- BioMysteryBench 90.3
- ARC-AGI-2 90
- SWE-bench Multilingual 89.1
- DRACO 87.7
- ProgramBench tests passed 87.6
- GDP.pdf 85.4
- SWE-Bench Pro 81.2
- OfficeQA 80.2
- OSWorld 2.0 77.9
- Toolathlon-Verified 77.8
- OfficeQA Pro 69
- DeepSWE 67.4
- Humanity's Last Exam 65
- HealthBench Professional 62.1
- HealthBench 60
- FrontierSWE 57
- Terminal-Bench 4.0 55.8
- SWE-bench Multimodal 54.7
- Terminal-Bench-Science 52.6
- FrontierCode 1.1 50.9
- BenchCAD 43.7
- Chartography 42.6
- AutomationBench 31.4
- Harvey LAB (Legal Agent Benchmark) 19.1
Its lab's numbers next to 12 leaderboards
Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.
| Benchmark | Lab says | Board says | Gap |
|---|---|---|---|
| GDP.pdf | 85.4% GDP.pdf mean criteria pass rate, no tools, Claude Opus 4.7 judge; 85.1% with tools (8.14.4) | 49.4% Leaderboard, adaptive/max | +36 pts |
| Humanity's Last Exam | 65% Multidisciplinary reasoning Humanity's Last Exam (with tools) | 46.5% Leaderboard, xhigh | +19 pts |
| Terminal-Bench-Science | 52.6% Agentic scientific research Terminal-Bench-Science 0.1 | 40% Leaderboard, max, Claude Code | +13 pts |
| Harvey LAB (Legal Agent Benchmark) | 19.1% Legal Agent Benchmark (Harvey LAB), all-pass rate on 1,235 public problems, max effort, internal harness, mean criterion pass rate 90.81%; held-out set run by Artificial Analysis 16.7% all-pass at xhigh (8.15.2) | 6.67% Leaderboard | +12 pts |
| AutomationBench | 31.4% Business workflows AutomationBench | 22.4% Leaderboard, max | +9.0 pts |
| ProgramBench tests passed | 87.6% ProgramBench hidden test pass rate on 166 tasks, single agent, mini-swe-agent harness, no time limit (8.11.1) | 82.5% Leaderboard, xhigh, mini-SWE-agent | +5.1 pts |
| Chartography | 42.6% Chartography, no tools, adaptive thinking max effort, 5 runs; 86.2% with tools (8.14.1) | 46.2% Leaderboard, adaptive/max | −3.6 pts |
| Terminal-Bench 4.0 | 55.8% Terminal-Bench 4.0, Claude Code --bare mode, max effort, 15 trials per task; Table 8.1.A rounds to 56% (Mythos 5.1 61%) (8.6) | 57.9% Leaderboard, max, Claude Code | −2.1 pts |
| scBench (LatchBio SingleCellBench) | 61.9% LatchBio SingleCellBench, Mythos 5.1 | 60.5% Leaderboard, published v1; LSVP, Pi | +1.4 pts |
| FrontierSWE | 57% FrontierSWE v2, mean score 0.57 on a 0 to 1 scale, max effort, Proximal harness, 5 trials per task (8.5) | 56.3% Leaderboard, mean@5, proximus | +0.7 pts |
| ARC-AGI | 97.5% ARC-AGI-1 | 97.5% Leaderboard, max | Same |
| ARC-AGI-2 | 90% ARC-AGI-2 | 90% Leaderboard, max | Same |
Anthropic's models over time
Anthropic's models in color, against every other model on the Epoch Capabilities Index, by release date.
Source: Epoch Capabilities Index, Epoch AI (CC BY 4.0)
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.
Its system card
Every result in the longer documents the lab published with it. 49 of them aren't in the launch post.
Capabilities
- SWE-Bench Pro 81.2
- SWE-bench Multilingual 89.1
- SWE-bench Multimodal 54.7
- DeepSWE 67.4
- FrontierCode 1.1 50.9
- FrontierSWE 57
- Terminal-Bench 4.0 55.8
- Terminal-Bench-Science 52.6
- CursorBench 3.2 73.4
- CritPt
- ProgramBench tests passed 87.6
- Humanity's Last Exam 60.9
- DRACO 87.7
- Chartography 42.6
- BenchCAD 43.7
- OSWorld 2.0 (Aug 2026) 77.9
- GDP.pdf 85.4
- OfficeQA 80.2
- OfficeQA Pro 69
- Harvey LAB (Legal Agent Benchmark) 19.1
- GDPval-AA v2 1853
- AA-Briefcase 1694
- Toolathlon-Verified 77.8
- AutomationBench 31.4
- ARC-AGI 97.5
- ARC-AGI-2 90
- Global MMLU 94
- MILU 93
- ArXivMath 91.3
- BioMysteryBench 90.3
Health
- HealthBench 60
- HealthBench Professional 62.1
Safety
- Single-turn harmful request evaluations (Anthropic) 94.7
- Single-turn benign request evaluations (Anthropic) 0
- Multi-turn testing (Anthropic)
- Child safety evaluations (Anthropic)
- Suicide and self-harm evaluations (Anthropic)
- Disordered eating evaluations (Anthropic)
- Election integrity evaluations (Anthropic)
- Malicious use of Claude Code (Anthropic) 90.3
- Malicious computer use (Anthropic) 85.7
Robustness
- Rewind-attacker cyber adversarial robustness (Anthropic) 4.5
- Gray Swan Shade knowledge elicitation through chat 1
- Gray Swan IPI Arena benchmark 0.1
- Prompt injection robustness against adaptive attackers, coding (Anthropic) 9.26
- Prompt injection robustness against adaptive attackers, computer use (Anthropic) 0.07
- Prompt injection robustness against adaptive attackers, browser use (Anthropic) 2.64