- Agentic coding SWE-bench Pro 63.2
- Agentic coding Terminal-Bench 2.1 80.4
- Multidisciplinary reasoning Humanity's Last Exam (with tools) 57.4
- Computer use OSWorld-Verified 81.2
- Knowledge work GDPval-AA v2 1618
- BrowseComp (multi agent) 86.6
- FrontierCode v1 38.8
- AutomationBench 13.5
- Best of Legal Agent Benchmark Full Public Set / Legal Agent Benchmark Harvey's Held-Out Set 8.9
- HealthBench Professional 57.8
Claude Sonnet 5
Claude Sonnet 5 places in the top three on 11 benchmarks.
196
Benchmarks with a score
0
Ranked first, out of five or more models
11
In the top three
Anthropic Jun 2026
Claude Sonnet 5
Closed weights Mid-size
157.0
#26 of 271
Epoch Capabilities Index
- DeepSWE 53.9% #18 /28
- Terminal-Bench 4.0 12.4% #21 /23
- ZeroBench 13% #24 /66
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 | |
|---|---|---|---|---|
| Phare | 78.1% | #3 of 62 | Official leaderboard | |
| BullshitBench V2 | 80.8% | #9 of 122 | Official leaderboard | |
| Humanity's Last Exam | 57.4% | #16 of 147 | Model card | |
| BullshitBench | 65.5% | #14 of 107 | Official leaderboard | |
| LMArena (Elo) | 1,462 | #47 of 333 | Official leaderboard | |
| INCLUDE | 86.5% | #4 of 25 | Model card | |
| Longform Creative Writing (EQ-Bench) | 78.3 | #20 of 119 | Official leaderboard | |
| Continual Learning Exam | 40.2% | #2 of 10 | Official leaderboard | |
| Creative Writing v3 | 76.3 | #28 of 120 | Official leaderboard | |
| Document Arena | 1,476 | #12 of 37 | Official leaderboard | |
| SEA-HELM | 73.8 | #21 of 62 | Official leaderboard | |
| Judgemark v4 | 70.9 | #15 of 40 | Official leaderboard | |
| EQ-Bench 4 | 1,221 | #14 of 32 | Official leaderboard | |
| Extended NYT Connections | 75.1% | #45 of 97 | Official leaderboard | |
| SimpleQA Verified | 32.1% | #28 of 56 | Official leaderboard | |
| Global MMLU | 89% | #7 of 11 | Model card | |
| AA-Omniscience | 20% | #13 of 17 | Model card | |
| MILU | 89.3% | #8 of 10 | Model card | |
| FACTS Parametric V2 | 20.5% | #19 of 20 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ProofBench | 66% | #7 of 59 | Official leaderboard | |
| GPQA Diamond | 90.5% | #40 of 219 | Epoch AI | |
| DTBench | 87.5% | #34 of 178 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 94.7% | #44 of 202 | Epoch AI | |
| CritPt | 16.9% | #41 of 160 | Epoch AI | |
| ProofBench v1.1 | 77% | #16 of 51 | Official leaderboard | |
| SimpleBench | 60.6% | #36 of 100 | Official leaderboard | |
| Surface Evolver Bench | 60% | #10 of 26 | Epoch AI | |
| Organic Chemistry (Anthropic) | 81% | #2 of 5 | Model card | |
| FrontierMath Tiers 1-3 v2 | 65.6% | #36 of 89 | Epoch AI | |
| BTF-3 | 0.14 | #5 of 10 | Epoch AI | |
| FrontierMath Tier 4 v2 | 29.3% | #36 of 71 | Epoch AI | |
| LatchBio Capabilities v1.0 | 40.9% | #16 of 27 | Official leaderboard | |
| ForecastBench | 63.3 | #52 of 86 | Official leaderboard | |
| Verifying the Verifiers (Project APE CRED benchmark) | 77.5% | #15 of 23 | Official leaderboard | |
| SpatialBench (LatchBio) | 70% | #4 of 6 | Model card | |
| USAMO 2026 | 79.5% | #4 of 6 | Model card | |
| ArXivMath | 65.7% | #7 of 8 | Model card | |
| scBench (LatchBio SingleCellBench) | 56% | #6 of 6 | Model card | |
| ProteinGym | 37% | #3 of 3 | Model card | |
| Structural Biology (Anthropic) | 69% | #2 of 2 | Model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 85.2% | #6 of 150 | Model card | |
| WeirdML | 68.8% | #22 of 112 | Official leaderboard | |
| ALE-Bench | 1,463 | #26 of 121 | Epoch AI | |
| Vibe Code Bench | 81.3% | #27 of 115 | Official leaderboard | |
| GBAEval | 65.3% | #5 of 21 | Official leaderboard | |
| SWE-bench Multilingual | 78.3% | #17 of 70 | Model card | |
| SWE-Bench Pro | 63.2% | #19 of 77 | Model card | |
| AndroidBench | 76.2% | #9 of 32 | Official leaderboard | |
| Code Migration | 44.4% | #23 of 78 | Official leaderboard | |
| GSO-Bench | 37.3% | #9 of 30 | Official leaderboard | |
| WebDev Arena | 1,537 | #34 of 105 | Epoch AI | |
| Code Arena | 1,540 | #44 of 124 | Official leaderboard | |
| SWE-Bench Pro V2 Hard | 88.2% | #6 of 16 | Official leaderboard | |
| Senior SWE-Bench | 28% | #9 of 23 | Official leaderboard | |
| FrontierCode | 42.7% | #20 of 46 | Epoch AI | |
| SWE-rebench | 56.8% | #6 of 13 | Official leaderboard | |
| FrontierCode 1.1 | 42.7% | #4 of 8 | Official leaderboard | |
| Quadruped reinforcement learning task (Anthropic) | 19.9 | #4 of 8 | Model card | |
| Image-to-WebDev Arena | 1,540 | #30 of 55 | Official leaderboard | |
| SWE Refactor Bench | 8.50% | #5 of 8 | Official leaderboard | |
| DeepSWE | 53.9% | #18 of 28 | Official leaderboard | |
| IOI v2 (Vals) | 45% | #33 of 45 | Official leaderboard | |
| Novel compiler task (Anthropic) | 76.1% | #3 of 4 | Model card | |
| Time series forecasting task (Anthropic) | 5.80 | #7 of 9 | Model card | |
| SWE-Bench Pro V2 | 93.2% | #13 of 16 | Official leaderboard | |
| Vibe Code Bench 1-100 | 13.8% | #22 of 26 | Official leaderboard | |
| Next.js Agent Evals | 81% | #12 of 14 | Official leaderboard | |
| MLS-Bench | 31.4% | #13 of 15 | Official leaderboard | |
| CursorBench | 34.1% | #18 of 19 | Epoch AI | |
| SWE-bench Multimodal | 28.1% | #10 of 10 | Model card | |
| CursorBench 3.1 | 61.2% | #4 of 4 | Model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LisanBench | 5724.0 | #21 of 133 | Official leaderboard | |
| BrowseComp | 86.6% | #15 of 77 | Model card | |
| Long-Horizon-Terminal-Bench (LHTB) | 49.7% | #5 of 25 | Official leaderboard | |
| Game Arena Five in a Row | 865 | #4 of 20 | Official leaderboard | |
| OSWorld-Verified | 81.2% | #8 of 37 | Model card | |
| LLM Chess | 806 | #30 of 124 | Official leaderboard | |
| Mystery Game Puzzles | 35% | #22 of 80 | Epoch AI | |
| Game Arena Go | 1,126 | #10 of 36 | Official leaderboard | |
| Agent Arena | 0.03 | #16 of 53 | Official leaderboard | |
| Toolathlon-Verified | 71.6% | #8 of 25 | Official leaderboard | |
| Chess Puzzles | 45% | #29 of 83 | Epoch AI | |
| Game Arena Nine Men's Morris | 745 | #13 of 36 | Official leaderboard | |
| Game Arena Dots and Boxes | 945 | #15 of 37 | Official leaderboard | |
| Toolathlon | 54.3% | #15 of 37 | Model card | |
| Game Arena Checkers | 949 | #17 of 38 | Official leaderboard | |
| CEO-Bench | $64,459 | #9 of 19 | Official leaderboard | |
| Game Arena Four in a Row | 330 | #21 of 38 | Official leaderboard | |
| Game Arena Clobber | 248 | #20 of 36 | Official leaderboard | |
| ApprenticeBench | 16% | #10 of 18 | Official leaderboard | |
| LMArena Search Arena | 1,193 | #22 of 35 | Official leaderboard | |
| Game Arena Dark Hex | 201 | #24 of 38 | Official leaderboard | |
| Kaggle Game Arena | 256 | #25 of 39 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 188 | #24 of 36 | Official leaderboard | |
| Terminal-Bench 2.1 | 74.6% | #9 of 13 | Official leaderboard | |
| Game Arena Reversi | 256 | #25 of 36 | Official leaderboard | |
| Game Arena Poker | -7.27 | #23 of 31 | Official leaderboard | |
| Game Arena Coin Game | 236 | #27 of 35 | Official leaderboard | |
| Game Arena Chess | 585 | #31 of 39 | Official leaderboard | |
| Game Arena Lines of Action | 339 | #29 of 36 | Official leaderboard | |
| Terminal-Bench 3.0 | 14.6% | #9 of 11 | Official leaderboard | |
| Game Arena Word Art | 146 | #30 of 35 | Official leaderboard | |
| Terminal-Bench 4.0 | 12.4% | #21 of 23 | Official leaderboard | |
| Game Arena Bargaining | 87 | #34 of 36 | Official leaderboard | |
| Game Arena Word Association | 18 | #34 of 36 | Official leaderboard | |
| AutomationBench | 13.5% | #41 of 41 | Model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ProfBench | 60.4% | #3 of 72 | Official leaderboard | |
| MortgageTax | 70% | #4 of 86 | Official leaderboard | |
| TaxEval v2 | 75.6% | #15 of 129 | Official leaderboard | |
| CorpFin v2 | 67% | #15 of 119 | Official leaderboard | |
| RCM-Bench-1 | 91.2% | #3 of 20 | Official leaderboard | |
| Excel Modeling Benchmark | 66.3% | #22 of 76 | Official leaderboard | |
| Legal Research Bench | 41.8% | #26 of 88 | Official leaderboard | |
| LegalBench | 83.9% | #46 of 138 | Official leaderboard | |
| MedCode | 47.5% | #32 of 96 | Official leaderboard | |
| Vending-Bench 2 | $6,378 | #22 of 66 | Official leaderboard | |
| Vals Finance Agent v2 | 53.9% | #30 of 80 | Official leaderboard | |
| GDPval-AA v2 | 1,618 | #10 of 26 | Model card | |
| Public Benefits Bench | 66% | #20 of 50 | Official leaderboard | |
| Tax Agent Bench | 29.4% | #32 of 79 | Official leaderboard | |
| APEX-Agents | 54.5% | #23 of 56 | Epoch AI | |
| HealthBench | 58.7% | #13 of 30 | Model card | |
| Harvey LAB (Legal Agent Benchmark) | 5.00% | #38 of 80 | Official leaderboard | |
| OfficeQA | 73.3% | #8 of 12 | Model card | |
| Holdings-IndividualTaxBench | 71.5% | #9 of 13 | Official leaderboard | |
| HealthBench Professional | 57.8% | #18 of 25 | Model card | |
| MedScribe | 76% | #71 of 98 | Official leaderboard | |
| EEBench | 40.3% | #19 of 26 | Official leaderboard | |
| RealmPathology | 71.7% | #10 of 13 | Official leaderboard | |
| TaxCalcBench v2 | 6.00% | #21 of 26 | Official leaderboard | |
| AA-Briefcase | 1,393 | #5 of 6 | Model card | |
| OfficeQA Pro | 59.4% | #13 of 14 | Model card | |
| PolicyBench | 72.3% | #48 of 50 | Official leaderboard | |
| BenchCAD | 26.6% | #10 of 10 | Model card | |
| Real-World Finance v2 (Anthropic) | 1,219 | #2 of 2 | Model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| GDP.pdf | 67.5% | #3 of 15 | Model card | |
| Vision Arena | 1,262 | #37 of 136 | Official leaderboard | |
| ZeroBench | 13% | #24 of 66 | Official leaderboard | |
| ChartMuseum | 70.1% | #2 of 5 | Model card | |
| GeoBench | 74% | #22 of 48 | Official leaderboard | |
| ParseBench | 61.2% | #26 of 42 | Official leaderboard | |
| CharXiv | 77% | #42 of 67 | Model card | |
| VoxelBench | 1,515 | #37 of 56 | Official leaderboard | |
| ArtifactArena | 986 | #14 of 21 | Official leaderboard | |
| Blueprint-Bench 2 | 24.9% | #25 of 35 | Official leaderboard | |
| Humanity's Sixth Sense | 24.8% | #24 of 29 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MASK | 96.9% | #3 of 21 | Model card | |
| Malicious use of Claude Code (Anthropic) | 92.4% | #3 of 16 | Model card | |
| Human Pathogen Capabilities Test v2 (HPCT v2) | 65% | #14 of 71 | Official leaderboard | |
| Protocols (Anthropic life sciences) | 62% | #1 of 3 | Model card | |
| MetagenomicsBench | 42.3% | #9 of 26 | Official leaderboard | |
| Gray Swan Agent Red Teaming (ART) benchmark | 0.10% | #5 of 14 | Model card | |
| Dictatorship Eval | 83.5% | #8 of 20 | Official leaderboard | |
| BrowseSafe++ | 88.4% | #6 of 15 | Official leaderboard | |
| VariantBench | 39.8% | #11 of 27 | Official leaderboard | |
| DrugDiscoveryBench | 50% | #8 of 19 | Official leaderboard | |
| SHADE-Arena | 10% | #7 of 16 | Model card | |
| Prompt injection robustness against adaptive attackers, coding (Anthropic) | 0.31% | #6 of 13 | Model card | |
| TxBench-Oligo-Discovery | 36.6% | #14 of 28 | Official leaderboard | |
| Gray Swan IPI Arena benchmark | 0.60% | #5 of 10 | Model card | |
| Anthropic automated behavioral audit | 2.53 | #3 of 6 | Model card | |
| BioSecBench-Function | 33.8% | #16 of 27 | Official leaderboard | |
| scBench-Long | 15.9% | #16 of 27 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 49.3% | #17 of 28 | Official leaderboard | |
| Prompt injection robustness against adaptive attackers, browser use (Anthropic) | 0.93% | #8 of 13 | Model card | |
| EpiBench | 21.7% | #17 of 27 | Official leaderboard | |
| OncologyBench | 14.7% | #20 of 30 | Official leaderboard | |
| Prompt injection live bug bounty (Anthropic with Gray Swan) | 0.19% | #2 of 3 | Model card | |
| Anthropic political even-handedness eval | 96.4% | #9 of 13 | Model card | |
| Malicious computer use (Anthropic) | 84.7% | #9 of 13 | Model card | |
| SpatialBench-Long | 24.4% | #19 of 27 | Official leaderboard | |
| DistressBench | 68% | #30 of 42 | Official leaderboard | |
| DeepsecBench | 14.9 | #20 of 28 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 61.9% | #31 of 43 | Official leaderboard | |
| Prompt injection robustness against adaptive attackers, computer use (Anthropic) | 2.25% | #10 of 13 | Model card | |
| BioSecBench-Surveillance | 39% | #21 of 27 | Official leaderboard | |
| Single-turn harmful request evaluations (Anthropic) | 96.7% | #14 of 18 | Model card | |
| TxBench-Antibody-Discovery | 32% | #22 of 28 | Official leaderboard | |
| Virology Capabilities Test (VCT) | 45% | #22 of 28 | Model card | |
| BBQ (Bias Benchmark for QA) | 72.4% | #19 of 24 | Model card | |
| DNA Synthesis Screening Evasion | 50% | #4 of 5 | Model card | |
| Firefox 147 exploit evaluation (Anthropic) | 13.2% | #4 of 5 | Model card | |
| Anthropic undermining liberal democracy (behavioral audit metric) | 1.22 | #5 of 6 | Model card | |
| Binary Exploitation Benchmark (Anthropic, OSS-Fuzz based) | 54.5% | #5 of 6 | Model card | |
| CyberGym | 52.7% | #28 of 33 | Model card | |
| SRE-Bench | 0.38% | #28 of 33 | Official leaderboard | |
| Single-turn benign request evaluations (Anthropic) | 0.59% | #17 of 18 | Model card | |
| ExploitBench | 31% | #18 of 19 | Model card | |
| LinuxArena | 0% | #6 of 6 | Model card | |
| Uncritically reporting flawed results (Anthropic) | 8.60 | #3 of 3 | Model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Political Science LLM Benchmark | 69.7% | #2 of 29 | Official leaderboard | |
| LMCA | 50% | #26 of 139 | Epoch AI | |
| ACCoRD | 73.8 | #30 of 150 | Official leaderboard | |
| Conceptual Reasoning Index | 61.8 | #31 of 149 | Official leaderboard | |
| Vals Multimodal Index | 68.8% | #8 of 33 | Official leaderboard | |
| SAGE (Vals) | 48.9% | #24 of 81 | 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. 67 of them aren't in the launch post.
System Card: Claude Sonnet 5 30 Jun 2026
Capabilities
- SWE-bench Verified 85.2
- SWE-Bench Pro 63.2
- SWE-bench Multilingual 78.3
- SWE-bench Multimodal 28.1
- Terminal-Bench 2.1 80.4
- FrontierCode 38.8
- CursorBench 3.1 61.2
- USAMO 2026 79.5
- ArXivMath 65.7
- ProgramBench
- Humanity's Last Exam 43.2
- BrowseComp 84.7
- GDP.pdf 67.5
- OSWorld-Verified 81.2
- BenchCAD 26.6
- ChartMuseum 70.1
- CharXiv 77
- OfficeQA 73.3
- OfficeQA Pro 59.4
- Real-World Finance v2 (Anthropic) 1219
- Harvey LAB (Legal Agent Benchmark) 5.8
- GDPval-AA v2 1618
- Toolathlon 54.3
- AutomationBench 13.5
- AA-Briefcase 1393
- Global MMLU 89
- MILU 89.3
- INCLUDE 86.5
- BioMysteryBench
- SpatialBench (LatchBio) 70
- scBench (LatchBio SingleCellBench) 56
- Structural Biology (Anthropic) 69
- ProteinGym 37
- Organic Chemistry (Anthropic) 81
- Protocols (Anthropic life sciences) 62
Health
- HealthBench 58.7
- HealthBench Professional 57.8
Hallucination
Safety
- Single-turn harmful request evaluations (Anthropic) 96.7
- Single-turn benign request evaluations (Anthropic) 0.59
- 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) 92.4
- Malicious computer use (Anthropic) 84.7
Robustness
- Gray Swan Agent Red Teaming (ART) benchmark 0.1
- Gray Swan IPI Arena benchmark 0.6
- Prompt injection live bug bounty (Anthropic with Gray Swan) 0.19
- Prompt injection robustness against adaptive attackers, coding (Anthropic) 0.31
- Prompt injection robustness against adaptive attackers, computer use (Anthropic) 2.25
- Prompt injection robustness against adaptive attackers, browser use (Anthropic) 0.93