- Agentic coding Terminal-Bench 4.0 (xhigh effort) 66.4
- Agentic coding FrontierCode v1.1 (Main) 54.4
- Agentic coding CursorBench 4.0 57.8
- Knowledge work GDPval-AA v2.1 1846
- Business workflows AutomationBench 40
- Multidisciplinary reasoning Humanity's Last Exam (with tools) 67.7
- Agentic scientific research Terminal-Bench-Science 0.1 58.7
- Computer use OSWorld 2.1 (partial) 81.8
- Visual chart recognition Chartography (with tools) 89
- SWE-bench Pro 89.9
- SWE-bench Multilingual 93.9
- SWE-bench Multimodal 61.4
- HealthBench Professional 65.6
- AA-Briefcase v1.1 1822
Claude Opus 5.5
Claude Opus 5.5 ranks first on 73 of the 250 benchmarks it has scores on.
269
Benchmarks with a score
73
Ranked first, out of five or more models
150
In the top three
Anthropic Sep 2026
Claude Opus 5.5
Closed weights Flagship
168.4
#2 of 271
Epoch Capabilities Index
- AutomationBench 42.5% #4 /5
- Agents' Last Exam 43.4% #2 /12
- Terminal-Bench 4.0 64.8% #1 /23
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 | |
|---|---|---|---|---|
| LMArena (Elo) | 1,509 | #3 of 333 | Official leaderboard | |
| Humanity's Last Exam | 67.7% | #2 of 147 | Model card | |
| Longform Creative Writing (EQ-Bench) | 84.6 | #3 of 119 | Official leaderboard | |
| SimpleQA Verified | 72.2% | #5 of 84 | Epoch AI | |
| EQ-Bench 4 | 1,340 | #2 of 32 | Official leaderboard | |
| Global MMLU | 94.3% | #1 of 11 | Model card | |
| Creative Writing v3 | 79.9 | #11 of 120 | Official leaderboard | |
| Global MMLU Lite | 95.4% | #4 of 43 | Official leaderboard | |
| MILU | 93.1% | #1 of 10 | Model card | |
| FACTS Multimodal | 48.4% | #4 of 39 | Official leaderboard | |
| AA-Omniscience | 58% | #2 of 17 | Model card | |
| BullshitBench V2 | 69.2% | #20 of 122 | Official leaderboard | |
| HLE-Diamond | 60.9% | #3 of 18 | Official leaderboard | |
| FACTS Benchmark Suite | 64% | #7 of 40 | Official leaderboard | |
| FACTS Search V2 | 86.5% | #4 of 21 | Official leaderboard | |
| BullshitBench | 53.7% | #23 of 107 | Official leaderboard | |
| Extended NYT Connections | 88.5% | #26 of 97 | Official leaderboard | |
| FACTS Parametric | 57.9% | #15 of 45 | Official leaderboard | |
| AI Detection Benchmark | 98.4% | #1 of 3 | Official leaderboard | |
| FACTS Parametric V2 | 58.7% | #9 of 20 | Official leaderboard | |
| Vectara Hallucination Leaderboard (2025 dataset) | 9.90% | #59 of 104 | Official leaderboard | |
| FACTS Grounding v2 | 62.5% | #10 of 15 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| DTBench | 98.2% | #1 of 178 | Official leaderboard | |
| SimpleBench | 88.4% | #1 of 100 | Official leaderboard | |
| ProofBench | 100% | #2 of 83 | Epoch AI | |
| CritPt | 31.7% | #4 of 160 | Epoch AI | |
| ARC-AGI-2 | 93.3% | #3 of 95 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 91.2% | #3 of 89 | Epoch AI | |
| LatchBio Capabilities v1.0 | 58% | #1 of 27 | Official leaderboard | |
| ProofBench v1.1 | 100% | #2 of 51 | Official leaderboard | |
| OTIS Mock AIME 2024-2025 | 100% | #8 of 202 | Epoch AI | |
| FrontierMath Tier 4 v2 | 95% | #3 of 71 | Epoch AI | |
| ARC-AGI | 98.5% | #4 of 92 | Official leaderboard | |
| FrontierPhysics | 22.6% | #1 of 19 | Official leaderboard | |
| Prophet Arena | 86.0 | #1 of 18 | Official leaderboard | |
| MysteryMechanism | 49.5% | #2 of 29 | Official leaderboard | |
| xbench-ScienceQA | 76.4% | #2 of 29 | Official leaderboard | |
| Terminal-Bench-Science | 63.3% | #2 of 17 | Official leaderboard | |
| ArXivLean | 47.2% | #1 of 7 | Official leaderboard | |
| BixBench3 | 50.2 | #3 of 20 | Official leaderboard | |
| scBench (LatchBio SingleCellBench) | 61.2% | #1 of 6 | Model card | |
| GPQA Diamond | 90.6% | #38 of 219 | Epoch AI | |
| FrontierMath Erdős | 2.94% | #3 of 14 | Epoch AI | |
| CritPt v2 | 77.7% | #3 of 11 | Official leaderboard | |
| Leipzig Benchmark | 85% | #6 of 19 | Official leaderboard | |
| ArXivMath | 83.8% | #2 of 6 | Official leaderboard | |
| BrokenArXiv | 84.7% | #2 of 6 | Official leaderboard | |
| Morphology-to-molecule matching (Anthropic) | 34% | #1 of 3 | Model card | |
| Human Behavior Bench | 79.3 | #5 of 11 | Official leaderboard | |
| SpatialBench (LatchBio) | 72% | #3 of 6 | Model card | |
| De novo protein binder design (Anthropic) | 82.6% | #1 of 2 | Model card | |
| Biomedical image analysis (Anthropic) | 71.4% | #2 of 3 | Model card | |
| Medicinal Chemistry (Anthropic) | 63.5% | #2 of 3 | Model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Code Arena | 1,810 | #1 of 124 | Official leaderboard | |
| WebDev Arena | 1,820 | #1 of 105 | Epoch AI | |
| SWE-Bench Pro | 89.9% | #1 of 77 | Model card | |
| SWE-bench Multilingual | 93.9% | #1 of 70 | Model card | |
| Image-to-WebDev Arena | 1,753 | #1 of 55 | Official leaderboard | |
| FrontierCode | 54.6% | #1 of 46 | Epoch AI | |
| FrontierCode 2.0 | 45.5% | #1 of 26 | Official leaderboard | |
| ProgramBench | 18.5% | #1 of 25 | Official leaderboard | |
| Vals RSI Index | 0.37 | #1 of 25 | Official leaderboard | |
| Vibe Code Bench | 90.3% | #5 of 115 | Official leaderboard | |
| Senior SWE-Bench | 39.3% | #1 of 23 | Official leaderboard | |
| TasteVal | 230.4× | #1 of 20 | Official leaderboard | |
| Code Migration | 66.7% | #4 of 78 | Official leaderboard | |
| Brokk Power Ranking | 37.9 | #1 of 19 | Official leaderboard | |
| CursorBench | 57.8% | #1 of 19 | Epoch AI | |
| RSI-Exam | 53.6% | #1 of 18 | Official leaderboard | |
| ALE-Bench | 2,147 | #7 of 121 | Epoch AI | |
| FrontierCS 2 | 58.3% | #1 of 16 | Official leaderboard | |
| Code Migration v1.1 | 81% | #2 of 29 | Official leaderboard | |
| Vibe Code Bench 1-100 | 30.4% | #2 of 26 | Official leaderboard | |
| SWE-bench-Live | 71.7% | #1 of 13 | Official leaderboard | |
| IOI v2 (Vals) | 95.1% | #4 of 45 | Official leaderboard | |
| FrontierSWE | 62.3% | #2 of 21 | Epoch AI | |
| GameDevBench | 70.6% | #1 of 10 | Official leaderboard | |
| SWE-bench Multimodal | 61.4% | #1 of 10 | Model card | |
| Tencent WorkBuddy Bench | 81.6% | #1 of 10 | Official leaderboard | |
| Web Clone Bench | 34.2% | #1 of 9 | Official leaderboard | |
| PostTrainBench 1.2 | 43.8% | #2 of 17 | Official leaderboard | |
| WeirdML v3 | 31.2% | #3 of 25 | Official leaderboard | |
| MirrorCode | 78.9% | #4 of 28 | Epoch AI | |
| Next.js Agent Evals | 97% | #2 of 14 | Official leaderboard | |
| DeepRSI | 1,629 | #1 of 7 | Official leaderboard | |
| DeepSWE | 74.2% | #9 of 51 | Model card | |
| Android Bench 2.0 | 37.3% | #2 of 11 | Official leaderboard | |
| ScarfBench | 70.5% | #2 of 10 | Official leaderboard | |
| FrontierCode 1.1 | 54.4% | #3 of 12 | Model card | |
| ReviewBench | 73.3% | #1 of 4 | Official leaderboard | |
| SWE-sweep | 3.60% | #3 of 10 | Official leaderboard | |
| ViBench 2.0 | 48.5% | #3 of 9 | Official leaderboard | |
| CursorBench 4.0 | 57.8% | #1 of 3 | Model card | |
| RSI Bench | 35.2 | #3 of 8 | Official leaderboard | |
| SWE-Marathon v1.1 | 46.3% | #7 of 17 | Official leaderboard | |
| CoBench (Anthropic) | 55.8% | #1 of 1 | Model card |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Agent Arena | 0.21 | #1 of 53 | Official leaderboard | |
| LisanBench | 24029.0 | #3 of 133 | Official leaderboard | |
| LLM Chess | 1,575 | #4 of 124 | Official leaderboard | |
| Terminal-Bench 4.0 | 64.8% | #1 of 23 | Official leaderboard | |
| Chess Puzzles | 67% | #4 of 83 | Epoch AI | |
| Mystery Game Puzzles | 71% | #4 of 80 | Epoch AI | |
| Game Arena Dots and Boxes | 1,172 | #2 of 37 | Official leaderboard | |
| Hermes Bench | 76.7% | #1 of 17 | Official leaderboard | |
| Time Horizon Index: KSP | 91.3% | #1 of 13 | Official leaderboard | |
| WitnessBench | 46.5% | #1 of 12 | Official leaderboard | |
| EBR-bench | 71.4% | #3 of 28 | Epoch AI | |
| Game Arena Nine Men's Morris | 975 | #4 of 36 | Official leaderboard | |
| Game Arena Word Art | 1,570 | #4 of 35 | Official leaderboard | |
| Game Arena Go | 1,554 | #5 of 36 | Official leaderboard | |
| Game Arena Checkers | 1,115 | #6 of 38 | Official leaderboard | |
| Game Arena Bargaining | 239 | #6 of 36 | Official leaderboard | |
| Game Arena Clobber | 486 | #6 of 36 | Official leaderboard | |
| Game Arena Lines of Action | 723 | #6 of 36 | Official leaderboard | |
| CivBench | 1,593 | #3 of 18 | Official leaderboard | |
| o11y-bench | 93.8% | #1 of 6 | Official leaderboard | |
| OSWorld 2.1 | 81.8% | #1 of 5 | Model card | |
| Kaggle Game Arena | 398 | #8 of 39 | Official leaderboard | |
| CUA-bench | 14% | #2 of 9 | Official leaderboard | |
| Toolathlon-Verified | 77.8% | #3 of 13 | Model card | |
| DRACO | 87.4% | #3 of 11 | Model card | |
| LMArena Search Arena | 1,211 | #11 of 35 | Official leaderboard | |
| Game Arena Chess | 962 | #13 of 39 | Official leaderboard | |
| WANDR | 72.3% | #1 of 3 | Model card | |
| Game Arena Dark Hex | 283 | #13 of 38 | Official leaderboard | |
| Game Arena Reversi | 692 | #13 of 36 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 363 | #15 of 36 | Official leaderboard | |
| StarSkirmish Bench | 38.6% | #4 of 9 | Official leaderboard | |
| SpireClimb | 3.33% | #4 of 7 | Epoch AI | |
| SpireClimb StS2 A10 | 6.67% | #4 of 7 | Epoch AI | |
| Game Arena Coin Game | 246 | #25 of 35 | Official leaderboard | |
| SpireClimb StS1 A20 Heart | 0% | #5 of 7 | Epoch AI | |
| AutomationBench | 42.5% | #4 of 5 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedScribe | 91.4% | #1 of 98 | Official leaderboard | |
| Excel Modeling Benchmark | 75.9% | #2 of 76 | Official leaderboard | |
| Vals Finance Agent v2.1 | 67.8% | #1 of 35 | Official leaderboard | |
| EEBench | 75% | #1 of 26 | Official leaderboard | |
| PolicyBench | 94.1% | #2 of 50 | Official leaderboard | |
| Tax Agent Bench | 45.6% | #4 of 79 | Official leaderboard | |
| APEX-Agents | 73.5% | #3 of 56 | Epoch AI | |
| Physical-Engineering's Last Exam | 0.49 | #1 of 18 | Official leaderboard | |
| Vals Finance Index | 65.6% | #2 of 35 | Official leaderboard | |
| Legora BAR | 106.0 | #1 of 17 | Official leaderboard | |
| Remote Labor Index | 21.3% | #1 of 17 | Official leaderboard | |
| Holdings-IndividualTaxBench | 76.3% | #1 of 13 | Official leaderboard | |
| Legal Research Bench | 50.5% | #7 of 88 | Official leaderboard | |
| Public Benefits Bench | 70.6% | #4 of 50 | Official leaderboard | |
| Immigration Bench | 72.1% | #2 of 24 | Official leaderboard | |
| Parametric CAD Bench | 61% | #1 of 11 | Official leaderboard | |
| Vals Index v2.2 | 70.2% | #3 of 32 | Official leaderboard | |
| Excel Modeling Benchmark v1.1 | 75.8% | #4 of 37 | Official leaderboard | |
| TaxCalcBench v2 | 70% | #3 of 26 | Official leaderboard | |
| Vals Finance Agent v2 | 58.6% | #10 of 80 | Official leaderboard | |
| GDPval-AA v2.1 | 1,846 | #1 of 8 | Model card | |
| Thrive Holdings helpdesk ticket resolution benchmark | 59.6% | #2 of 15 | Official leaderboard | |
| Clipping Bench | $2,018 | #3 of 19 | Official leaderboard | |
| HealthBench Professional | 65.6% | #4 of 25 | Model card | |
| Agents' Last Exam | 43.4% | #2 of 12 | Official leaderboard | |
| CAD Arena | 0.75 | #2 of 12 | Official leaderboard | |
| AA-Briefcase v1.1 | 1,822 | #1 of 6 | Model card | |
| Vending-Bench 2 | $9,235 | #12 of 66 | Official leaderboard | |
| RWE-bench | 29% | #2 of 11 | Official leaderboard | |
| MedCode | 49.8% | #19 of 96 | Official leaderboard | |
| Vending-Bench Arena | $8,094 | #7 of 32 | Official leaderboard | |
| RealmPathology | 77% | #3 of 13 | Official leaderboard | |
| Vals Legal Index | 72.3% | #8 of 33 | Official leaderboard | |
| FrontierFinance | 50.9% | #4 of 16 | Official leaderboard | |
| OfficeQA | 78.9% | #3 of 12 | Model card | |
| OfficeQA Pro | 67.7% | #4 of 14 | Model card | |
| Ivo-micro1 ContractBench | 58.1% | #2 of 7 | Official leaderboard | |
| HealthBench | 60.6% | #9 of 30 | Model card | |
| MissionReport | 63.5% | #3 of 9 | Official leaderboard | |
| RCM-Bench-1 | 90.1% | #7 of 20 | Official leaderboard | |
| LegalReasonBench | 46.5% | #3 of 8 | Official leaderboard | |
| Quant-Arena | 44.3 | #4 of 9 | Official leaderboard | |
| PRBench | 47.6% | #20 of 44 | Official leaderboard | |
| micro1 Medical Index | 68.4% | #4 of 8 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 3.75% | #42 of 80 | Official leaderboard | |
| BenchCAD | 73% | #6 of 10 | Model card | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 9.17% | #21 of 33 | Official leaderboard | |
| PharmaBench (micro1) | 10.8% | #8 of 12 | Official leaderboard | |
| Short Form Bench | $2,389 | #6 of 9 | Official leaderboard |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ParseBench | 78.9% | #1 of 42 | Official leaderboard | |
| Furniture Assembly Benchmark | 83.3% | #1 of 35 | Epoch AI | |
| ZeroBench | 43% | #2 of 66 | Official leaderboard | |
| Code2Video Benchmark | 1,585 | #1 of 19 | Official leaderboard | |
| Drone-Bench | 97.3% | #1 of 19 | Official leaderboard | |
| VoxelBench | 2,605 | #3 of 56 | Official leaderboard | |
| ClockBench | 77.2% | #2 of 37 | Official leaderboard | |
| Vision Arena | 1,293 | #8 of 136 | Official leaderboard | |
| Blueprint-Bench 2 | 51.2% | #3 of 35 | Official leaderboard | |
| RoboColiseum | 76.3% | #1 of 9 | Official leaderboard | |
| Humanity's Sixth Sense | 44.6% | #4 of 29 | Official leaderboard | |
| GDP.pdf | 30.6% | #6 of 39 | Epoch AI | |
| Chartography | 89% | #2 of 12 | Model card | |
| VibeVideoBench | 4.07 | #2 of 9 | Official leaderboard | |
| Embodied First Exam | 23.5% | #2 of 7 | Official leaderboard | |
| ScienceVision | 0.45 | #4 of 11 | Official leaderboard | |
| Robo-bench | 14.6% | #3 of 7 | Official leaderboard | |
| DataCenterBench Tier 1 | 21.7% | #4 of 9 | Official leaderboard | |
| RoboDojo-RC Tier 1 | 38.2% | #4 of 9 | Official leaderboard | |
| OmniParseBench | 90.0 | #3 of 6 | Official leaderboard | |
| Dual-Arm Sim | 19.0 | #7 of 10 | Official leaderboard | |
| LADDER | 49% | #3 of 4 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| FORTRESS | 5.78% | #1 of 66 | Official leaderboard | |
| OncologyBench | 50.7% | #1 of 30 | Official leaderboard | |
| TxBench-Antibody-Discovery | 67.7% | #1 of 28 | Official leaderboard | |
| TxBench-Oligo-Discovery | 62.2% | #1 of 28 | Official leaderboard | |
| TxBench-Preclinical-Pharmacology | 65.3% | #1 of 28 | Official leaderboard | |
| BioSecBench-Function | 73.6% | #1 of 27 | Official leaderboard | |
| BioSecBench-Surveillance | 60.8% | #1 of 27 | Official leaderboard | |
| EpiBench | 39.6% | #1 of 27 | Official leaderboard | |
| scBench-Long | 36.9% | #1 of 27 | Official leaderboard | |
| SpatialBench-Long | 40.5% | #1 of 27 | Official leaderboard | |
| MetagenomicsBench | 62.7% | #1 of 26 | Official leaderboard | |
| PCT (Polarized Contrastive Pairs) | 29.1% | #2 of 46 | Official leaderboard | |
| SHADE-Arena | 26.2% | #1 of 16 | Model card | |
| Anthropic political even-handedness eval | 99.4% | #1 of 13 | Model card | |
| VariantBench | 50.6% | #3 of 27 | Official leaderboard | |
| SRE-Bench | 33.6% | #4 of 33 | Official leaderboard | |
| CyScenarioBench | 67.6% | #2 of 14 | Model card | |
| ExploitGym | 34.5% | #3 of 18 | Official leaderboard | |
| Binary Exploitation Benchmark (Anthropic, OSS-Fuzz based) | 91.1% | #1 of 6 | Model card | |
| Impossible tasks (Anthropic) | 34.7% | #2 of 11 | Model card | |
| CheatBench | 11.2% | #2 of 10 | Official leaderboard | |
| Rewind-attacker cyber adversarial robustness (Anthropic) | 4.00% | #1 of 5 | Model card | |
| ExploitBench | 91% | #4 of 19 | Model card | |
| Prompt injection robustness against adaptive attackers, browser use (Anthropic) | 0.09% | #3 of 13 | Model card | |
| Prompt injection robustness against adaptive attackers, computer use (Anthropic) | 0.07% | #3 of 13 | Model card | |
| Virology Capabilities Test (VCT) | 59% | #7 of 28 | Model card | |
| BBQ (Bias Benchmark for QA) | 89.7% | #6 of 24 | Model card | |
| Self-preference bias (Anthropic) | 0.07 | #1 of 4 | Model card | |
| DistressBench | 75.6% | #11 of 42 | Official leaderboard | |
| Single-turn benign request evaluations (Anthropic) | 0.03% | #6 of 18 | Model card | |
| Anthropic undermining liberal democracy (behavioral audit metric) | 1.17 | #2 of 6 | Model card | |
| Concealment of changes (Anthropic) | 96.9% | #1 of 3 | Model card | |
| Sandbox escape evaluation (Anthropic) | 1.50% | #1 of 3 | Model card | |
| Training distribution hack rate (Anthropic) | 0.63% | #1 of 3 | Model card | |
| DeepsecBench | 26.6 | #10 of 28 | Official leaderboard | |
| CWE-bench | 67% | #5 of 12 | Official leaderboard | |
| BrowseSafe++ | 85.3% | #7 of 15 | Official leaderboard | |
| LinuxArena | 33.4% | #3 of 6 | Model card | |
| Cyber harm coverage (Anthropic) | 99.7% | #2 of 4 | Model card | |
| MASK | 87.4% | #12 of 21 | Model card | |
| WellnessBench | 53% | #8 of 12 | Official leaderboard | |
| Anthropic automated behavioral audit | 2.49 | #4 of 6 | Model card | |
| Silent copying of answers (Anthropic) | 12.1% | #2 of 3 | Model card | |
| Malicious use of Claude Code (Anthropic) | 79.8% | #13 of 16 | Model card | |
| CyberBench v1.1 (Vals) | 55.4% | #35 of 43 | Official leaderboard | |
| BioSecBench-Refusal | 3.80% | #18 of 22 | Official leaderboard | |
| Gray Swan IPI Arena benchmark | 0.10% | #9 of 10 | Model card | |
| Malicious computer use (Anthropic) | 79.5% | #12 of 13 | Model card | |
| Single-turn harmful request evaluations (Anthropic) | 94.5% | #18 of 18 | Model card | |
| Prompt injection robustness against adaptive attackers, coding (Anthropic) | 54.6% | #13 of 13 | Model card | |
| Compliance with instructions planted in pasted text (Anthropic) | 7.40% | #1 of 1 | Model card | |
| Destructive actions (Anthropic) | 21% | #1 of 1 | Model card | |
| Package registry publishing evaluation (Anthropic) | 33.3% | #1 of 1 | Model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ACCoRD | 86.5 | #1 of 150 | Official leaderboard | |
| Conceptual Reasoning Index | 77.9 | #1 of 149 | Official leaderboard | |
| LMCA | 68.2% | #2 of 139 | Epoch AI | |
| CAIS Risk Index | 26.2 | #1 of 65 | Official leaderboard | |
| SAGE v1.1 (Vals) | 91.3% | #1 of 28 | Official leaderboard | |
| Hermes Index | 63.3 | #1 of 17 | Official leaderboard | |
| CAIS Text Capabilities Index | 56.3 | #4 of 66 | Official leaderboard | |
| CAIS Vision Capabilities Index | 76.8 | #4 of 53 | Official leaderboard | |
| LiveBench | 85.1% | #3 of 28 | Official leaderboard | |
| Anthropic ECI (AECI) | 169.4 | #1 of 6 | Model card | |
| Vals Science Index | 58.6% | #4 of 23 | Official leaderboard | |
| SAGE (Vals) | 45.8% | #37 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. 74 of them aren't in the launch post.
System Card: Claude Opus 5.5 22 Sep 2026
Capabilities
- SWE-Bench Pro 89.9
- SWE-bench Multilingual 93.9
- SWE-bench Multimodal 61.4
- DeepSWE 74.2
- FrontierCode 1.1 54.4
- Terminal-Bench 4.0 66.4
- Terminal-Bench-Science 58.7
- FrontierSWE 62.3
- CursorBench 4.0 57.8
- ArXivMath 91.2
- ProgramBench 91.2
- Humanity's Last Exam 64.4
- DRACO 87.4
- WANDR 72.3
- Anthropic large agent teams tasks
- Chartography 64.4
- BenchCAD 73
- OSWorld 2.1 81.8
- OfficeQA 78.9
- OfficeQA Pro 67.7
- Harvey LAB (Legal Agent Benchmark) 8.3
- GDPval-AA v2.1 1846
- AA-Briefcase v1.1 1822
- Toolathlon-Verified 77.8
- AutomationBench 40
- Global MMLU 94.3
- MILU 93.1
- SpatialBench (LatchBio) 72
- scBench (LatchBio SingleCellBench) 61.2
- Morphology-to-molecule matching (Anthropic) 34
- Medicinal Chemistry (Anthropic) 63.5
- Protein Design (Anthropic)
- De novo protein binder design (Anthropic) 82.6
- Biomedical image analysis (Anthropic) 71.4
Health
- HealthBench 60.6
- HealthBench Professional 65.6
Hallucination
Safety
- Single-turn harmful request evaluations (Anthropic) 94.5
- Single-turn benign request evaluations (Anthropic) 0.03
- 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) 79.8
- Malicious computer use (Anthropic) 79.5
Robustness
- Rewind-attacker cyber adversarial robustness (Anthropic) 4
- Gray Swan IPI Arena benchmark 0.1
- Prompt injection robustness against adaptive attackers, coding (Anthropic) 54.6
- Prompt injection robustness against adaptive attackers, computer use (Anthropic) 0.07
- Prompt injection robustness against adaptive attackers, browser use (Anthropic) 0.09
Alignment
- Training distribution hack rate (Anthropic) 0.63
- Impossible tasks (Anthropic) 34.7
- Anthropic automated behavioral audit 2.49
- Anthropic undermining liberal democracy (behavioral audit metric) 1.17
- Sandbox escape evaluation (Anthropic) 1.5
- Package registry publishing evaluation (Anthropic) 33.3
- Compliance with instructions planted in pasted text (Anthropic) 7.4
- Destructive actions (Anthropic) 21
- Self-preference bias (Anthropic) 0.07
- MASK 87.4
- Silent copying of answers (Anthropic) 12.1
- Concealment of changes (Anthropic) 96.9
Monitorability
Bio and chem risk
Cyber risk
AI R&D risk
- CoBench (Anthropic) 55.8
- Anthropic ECI (AECI) 169.4