- Agentic coding SWE-Bench Pro 80.3
- Agentic coding FrontierCode (Diamond) (xhigh) 29.3
- Knowledge work GDPval-AA 1932
- Knowledge work vision GDP.pdf (no tools) 29.8
- Spatial reasoning Blueprint-Bench 2 38.6
- Tool use AutomationBench 17.4
- Computer use OSWorld-Verified 85
- Best of Legal Legal Agent Benchmark (Harvey's Held-Out Set) / Legal Agent Full Public Benchmark Set 16.9
- Multidisciplinary reasoning Humanity's Last Exam (with tools) 64.5
- Best of Biology BioMysteryBench (hard) / Biology BioMysteryBench (human solved) 83.9
- Agentic coding Terminal-Bench 2.1 88
- Health HealthBench Professional 66
- HealthBench 62.7
- SWE-bench Verified 95.5
- BrowseComp (multi-agent) 93.3
- CharXiv Reasoning (with tools) 93.5
- CritPt 28.6
- ArxivMath 78.5
- RiemannBench 55
- Best of GraphWalks BFS 256K / GraphWalks Parents 256K 100
- OfficeQA Pro 57.9
Claude Fable 5
Claude Fable 5 ranks first on 24 of the 161 benchmarks it has scores on.
168
Benchmarks with a score
24
Ranked first, out of five or more models
55
In the top three
Anthropic Jun 2026
Claude Fable 5
Closed weights Flagship
163.0
#11 of 271
Epoch Capabilities Index
- DeepSWE 69.9% #5 /28
- Terminal-Bench 4.0 44.5% #8 /23
- τ³-bench Banking 39.7% #9 /25
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 | |
|---|---|---|---|---|
| EQ-Bench 3 | 2,096 | #1 of 78 | Official leaderboard | |
| LMArena (Elo) | 1,504 | #6 of 333 | Official leaderboard | |
| Creative Writing v3 | 81.9 | #3 of 120 | Official leaderboard | |
| Document Arena | 1,499 | #1 of 37 | Official leaderboard | |
| Longform Creative Writing (EQ-Bench) | 83.0 | #4 of 119 | Official leaderboard | |
| Humanity's Last Exam | 64.5% | #6 of 147 | Model card | |
| GraphWalks | 100% | #1 of 21 | Model card | |
| BullshitBench V2 | 83.6% | #7 of 122 | Official leaderboard | |
| SimpleQA Verified | 70.7% | #7 of 84 | Epoch AI | |
| Extended NYT Connections | 92.7% | #15 of 97 | Official leaderboard | |
| EQ-Bench 4 | 1,313 | #5 of 32 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ARC-AGI | 98.5% | #1 of 92 | Official leaderboard | |
| DTBench | 97.3% | #2 of 178 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 100% | #3 of 202 | Epoch AI | |
| ProofBench | 77% | #2 of 59 | Official leaderboard | |
| Surface Evolver Bench | 95% | #1 of 26 | Epoch AI | |
| SimpleBench | 81.9% | #7 of 100 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 90.2% | #5 of 71 | Epoch AI | |
| CritPt | 28.6% | #13 of 160 | Epoch AI | |
| ARC-AGI-2 | 89.2% | #9 of 95 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 87% | #10 of 89 | Epoch AI | |
| xbench-ScienceQA | 73.4% | #4 of 29 | Official leaderboard | |
| ProofBench v1.1 | 95% | #8 of 51 | Official leaderboard | |
| Verifying the Verifiers (Project APE CRED benchmark) | 96.5% | #6 of 23 | Official leaderboard | |
| PaperBenchX | 11.8% | #3 of 10 | Official leaderboard | |
| GPQA Diamond | 85.9% | #69 of 219 | Epoch AI | |
| Terminal-Bench-Science | 21.4% | #6 of 17 | Official leaderboard | |
| BTF-3 | 0.13 | #4 of 10 | Epoch AI | |
| ArXivMath | 78.5% | #4 of 8 | Model card | |
| BioMysteryBench | 83.9% | #4 of 8 | Model card | |
| FrontierMath Erdős | 0% | #8 of 14 | Epoch AI | |
| Prophet Arena | 84.8 | #14 of 18 | Official leaderboard | |
| ForecastBench | 65.6 | #78 of 86 | Official leaderboard | |
| Riemann-bench | 55% | #4 of 4 | Model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 95.5% | #2 of 150 | Model card | |
| OpenHands Index | 81% | #1 of 33 | Official leaderboard | |
| Vibe Code Bench | 90.3% | #4 of 115 | Official leaderboard | |
| WeirdML | 92% | #4 of 112 | Official leaderboard | |
| FrontierCode | 53.5% | #2 of 46 | Epoch AI | |
| SWE-Bench Pro | 80.3% | #4 of 77 | Model card | |
| AndroidBench | 90.9% | #2 of 32 | Official leaderboard | |
| SWE-rebench | 64.5% | #1 of 13 | Official leaderboard | |
| ALE-Bench | 2,041 | #10 of 121 | Epoch AI | |
| IOI (Vals) | 72.3% | #5 of 55 | Official leaderboard | |
| GBAEval | 74.5% | #2 of 21 | Official leaderboard | |
| GSO-Bench | 78.4% | #3 of 30 | Official leaderboard | |
| Vision2Web | 82.3% | #1 of 10 | Official leaderboard | |
| WebDev Arena | 1,628 | #13 of 105 | Epoch AI | |
| FrontierCode 1.1 | 53.5% | #1 of 8 | Official leaderboard | |
| Code Migration | 55.1% | #10 of 78 | Official leaderboard | |
| PostTrainBench | 30.7% | #3 of 23 | Official leaderboard | |
| Senior SWE-Bench | 34.7% | #3 of 23 | Official leaderboard | |
| SWE Atlas Test Writing | 55.6% | #3 of 22 | Official leaderboard | |
| SWE Atlas Refactoring | 54.8% | #3 of 17 | Official leaderboard | |
| DeepSWE | 69.9% | #5 of 28 | Official leaderboard | |
| Image-to-WebDev Arena | 1,623 | #10 of 55 | Official leaderboard | |
| PostTrainBench 1.1 | 41.8% | #2 of 11 | Official leaderboard | |
| Code Arena | 1,624 | #23 of 124 | Official leaderboard | |
| MLS-Bench | 49.9% | #3 of 15 | Official leaderboard | |
| MirrorCode | 63.9% | #7 of 28 | Epoch AI | |
| SWE-Marathon | 24% | #3 of 12 | Official leaderboard | |
| CursorBench 3.1 | 72.9% | #1 of 4 | Model card | |
| FrontierSWE | 47% | #7 of 21 | Epoch AI | |
| PostTrainBench 1.2 | 37.3% | #6 of 17 | Official leaderboard | |
| Vals RSI Index | 0.24 | #10 of 25 | Official leaderboard | |
| WeirdML v3 | 15.6% | #10 of 25 | Official leaderboard | |
| GameDevBench | 67.3% | #4 of 10 | Official leaderboard | |
| SWE Atlas | 39% | #12 of 22 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BrowseComp | 93.3% | #1 of 77 | Model card | |
| Pencil Puzzle Bench | 97.6% | #1 of 53 | Official leaderboard | |
| OSWorld-Verified | 86% | #1 of 33 | Official leaderboard | |
| LisanBench | 13680.3 | #10 of 133 | Official leaderboard | |
| PACT | 1,603 | #2 of 26 | Official leaderboard | |
| CEO-Bench | $12,630,078 | #2 of 19 | Official leaderboard | |
| Mystery Game Puzzles | 52% | #12 of 80 | Epoch AI | |
| Terminal-Bench 2.1 | 83.8% | #2 of 13 | Official leaderboard | |
| ApprenticeBench | 43% | #3 of 18 | Official leaderboard | |
| LMArena Search Arena | 1,229 | #6 of 35 | Official leaderboard | |
| Agent Arena | 0.08 | #10 of 53 | Official leaderboard | |
| MCP Atlas | 83.3% | #7 of 34 | Official leaderboard | |
| Chess Puzzles | 47% | #22 of 83 | Epoch AI | |
| Toolathlon | 61.7% | #10 of 37 | Model card | |
| Terminal-Bench 3.0 | 34% | #3 of 11 | Official leaderboard | |
| Long-Horizon-Terminal-Bench (LHTB) | 48.7% | #7 of 25 | Official leaderboard | |
| Terminal-Bench 4.0 | 44.5% | #8 of 23 | Official leaderboard | |
| τ³-bench Banking | 39.7% | #9 of 25 | Official leaderboard | |
| EBR-bench | 39.5% | #11 of 28 | Epoch AI | |
| Vals Web Search Index | 48.5% | #2 of 4 | Official leaderboard | |
| PinchBench 2.0 | 54.8% | #44 of 57 | Official leaderboard | |
| AutomationBench | 17.4% | #37 of 41 | Model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LegalBench | 88.6% | #1 of 138 | Official leaderboard | |
| CorpFin v2 | 71.8% | #2 of 119 | Official leaderboard | |
| TaxEval v2 | 76.9% | #5 of 129 | Official leaderboard | |
| MedCode | 56.1% | #4 of 96 | Official leaderboard | |
| GDPval-AA | 1,932 | #1 of 16 | Model card | |
| Excel Modeling Benchmark | 73.7% | #5 of 76 | Official leaderboard | |
| RadLE 2.0 | 758.0 | #1 of 12 | Official leaderboard | |
| CADGenBench | 45.1% | #1 of 11 | Official leaderboard | |
| Public Benefits Bench | 70.4% | #5 of 50 | Official leaderboard | |
| Legal Research Bench | 49.5% | #9 of 88 | Official leaderboard | |
| MortgageTax | 68.9% | #9 of 86 | Official leaderboard | |
| E-Commerce Bench | 8.05 | #2 of 19 | Official leaderboard | |
| PRBench | 52.6% | #5 of 44 | Official leaderboard | |
| HealthBench Professional | 66% | #3 of 25 | Model card | |
| MedScribe | 88.5% | #12 of 98 | Official leaderboard | |
| JobBench | 57.4% | #2 of 16 | Official leaderboard | |
| Tax Agent Bench | 38.3% | #12 of 79 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 11.3% | #14 of 80 | Official leaderboard | |
| HealthBench | 62.7% | #6 of 30 | Model card | |
| prinzbench | 56.0 | #6 of 26 | Official leaderboard | |
| APEX-Agents | 63.6% | #13 of 56 | Epoch AI | |
| Vals Finance Agent v2 | 56.3% | #19 of 80 | Official leaderboard | |
| Remote Labor Index | 15.8% | #5 of 17 | Official leaderboard | |
| Physical-Engineering's Last Exam | 0.34 | #6 of 18 | Official leaderboard | |
| Real-World Finance (Anthropic) | 70% | #1 of 3 | Model card | |
| FrontierFinance | 49.2% | #6 of 16 | Official leaderboard | |
| Vending-Bench 2 | $5,680 | #27 of 66 | Official leaderboard | |
| TaxCalcBench v2 | 44% | #12 of 26 | Official leaderboard | |
| PolicyBench | 83.3% | #24 of 50 | Official leaderboard | |
| RedlineBench | 47.3% | #2 of 4 | Official leaderboard | |
| Real-World Finance v2 (Anthropic) | 1,374 | #1 of 2 | Model card | |
| EEBench | 54.3% | #14 of 26 | Official leaderboard | |
| Vending-Bench Arena | $4,244 | #18 of 32 | Official leaderboard | |
| OfficeQA | 57.9% | #11 of 12 | Model card | |
| OfficeQA Pro | 57.9% | #14 of 14 | Model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vision Arena | 1,309 | #1 of 136 | Official leaderboard | |
| CharXiv | 93.5% | #1 of 67 | Model card | |
| GeoBench | 89% | #1 of 48 | Official leaderboard | |
| BabyVision | 90.5% | #2 of 27 | Official leaderboard | |
| ZeroBench | 24% | #6 of 66 | Official leaderboard | |
| VoxelBench | 2,120 | #10 of 56 | Official leaderboard | |
| GDP.pdf | 30% | #7 of 39 | Epoch AI | |
| ArtifactArena | 1,189 | #4 of 21 | Official leaderboard | |
| EnigmaEval | 39.3% | #1 of 5 | Official leaderboard | |
| Blueprint-Bench 2 | 38.6% | #8 of 35 | Official leaderboard | |
| ParseBench | 69.8% | #11 of 42 | Official leaderboard | |
| Drone-Bench | 84.3% | #5 of 19 | Official leaderboard | |
| Humanity's Sixth Sense | 34.5% | #10 of 29 | Official leaderboard | |
| ClockBench | 37.2% | #14 of 37 | Official leaderboard | |
| Furniture Assembly Benchmark | 35.8% | #21 of 35 | Epoch AI |
Safety and security
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ACCoRD | 85.0 | #3 of 150 | Official leaderboard | |
| Vals Multimodal Index | 74.2% | #1 of 33 | Official leaderboard | |
| Conceptual Reasoning Index | 72.7 | #7 of 149 | Official leaderboard | |
| LMCA | 61.1% | #8 of 139 | Epoch AI | |
| CAIS Risk Index | 30.8 | #4 of 65 | Official leaderboard | |
| CAIS Text Capabilities Index | 54.4 | #6 of 66 | Official leaderboard | |
| SAGE (Vals) | 51.9% | #9 of 81 | Official leaderboard | |
| CAIS Vision Capabilities Index | 66.6 | #10 of 53 | Official leaderboard |
Anthropic's models over time
Anthropic's models in color, against every other model on the Epoch Capabilities Index, by release date.
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. 31 of them aren't in the launch post.
Capabilities
- SWE-Bench Pro 80
- SWE-bench Verified 95
- Terminal-Bench 2.1 84.3
- FrontierCode 29.3
- FrontierSWE v1
- CursorBench 3.1 72.9
- OSWorld-Verified 85
- GDP.pdf 29.8
- Blueprint-Bench 2 38.6
- LAB-Bench FigQA
- OfficeQA Pro 57.9
- Vals Finance Agent v2 56.3
- Real-World Finance v2 (Anthropic) 1374
- Real-World Finance (Anthropic) 70
- Harvey LAB (Legal Agent Benchmark) 13.3
- MCP Atlas 83.3
- Vending-Bench 2 5680.3
- GDPval-AA 1932
- Toolathlon 61.7
- AutomationBench 17.4
Safety
- Single-turn harmful request evaluations (Anthropic) 96.9
- Single-turn benign request evaluations (Anthropic) 0.01
- 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) 5
- Gray Swan Agent Red Teaming (ART) benchmark 0.1
- Prompt injection robustness against adaptive attackers, coding (Anthropic) 0.45
- Prompt injection robustness against adaptive attackers, computer use (Anthropic) 0.82
- Prompt injection robustness against adaptive attackers, browser use (Anthropic) 29.7