- Knowledge work GDPval-AA v2.1 1620
- Knowledge work AA-Briefcase v1.1 1578
- Computer use OSWorld 2.1 (Offline subset) 72.4
- Multidisciplinary reasoning Humanity's Last Exam (with tools) 57.4
- Agentic coding Terminal-Bench 4.0 39.2
- Agentic coding FrontierCode 1.1 (Main) (max) 46.4
- Visual reasoning Chartography (no tools) 46.4
- SWE-Bench Pro 64.8
- SWE-bench Multilingual 83.7
- SWE-bench Multimodal 30.7
- HealthBench Professional (Length-adjusted) 64.8
Claude Haiku 5.5
Claude Haiku 5.5 places in the top three on 5 benchmarks.
211
Benchmarks with a score
0
Ranked first, out of five or more models
5
In the top three
Anthropic Oct 2026
Claude Haiku 5.5
Closed weights Small
160.4
#18 of 256
Epoch Capabilities Index
Self-Reported Index 156.8 #50 of 404
- Humanity's Last Exam 30.1% #13 /47
- AutomationBench 12% #26 /42
- MultiChallenge 67.7% #10 /42
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 | |
|---|---|---|---|---|
| MultiChallenge | 67.7% | #10 of 42 | Official leaderboard | |
| Humanity's Last Exam | 30.1% | #13 of 47 | Official leaderboard | |
| BullshitBench V2 | 57.6% | #39 of 129 | Official leaderboard | |
| LMArena (Elo) | 1,422 | #102 of 335 | Official leaderboard | |
| DesignArena ASCII Art | 1,215 | #25 of 77 | Official leaderboard | |
| AA-LCR Extended | 63.8% | #23 of 49 | Official leaderboard | |
| Extended NYT Connections | 65.7% | #62 of 103 | Official leaderboard | |
| Global MMLU Lite | 87.3% | #26 of 43 | Official leaderboard | |
| SimpleQA Verified | 23.9% | #37 of 56 | Official leaderboard | |
| Global MMLU | 87.8% | #10 of 14 | Lab model card | |
| FACTS Search V2 | 77.5% | #16 of 21 | Official leaderboard | |
| FACTS Grounding v2 | 57% | #20 of 25 | Official leaderboard | |
| FACTS Parametric | 13.1% | #37 of 45 | Official leaderboard | |
| FACTS Multimodal | 38.9% | #20 of 23 | Official leaderboard | |
| MILU | 87.6% | #10 of 11 | Lab model card | |
| FACTS Parametric V2 | 13.1% | #25 of 27 | Official leaderboard | |
| FACTS Benchmark Suite | 47.3% | #23 of 23 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.9% | #28 of 212 | Epoch AI run | |
| DTBench | 95% | #24 of 180 | Official leaderboard | |
| GPQA Diamond | 89.7% | #52 of 229 | Epoch AI run | |
| ProofBench v1.1 | 89% | #12 of 52 | Official leaderboard | |
| FrontierMath Tiers 1-3 v2 | 75.1% | #22 of 86 | Official leaderboard | |
| Prophet Arena | 85.9 | #8 of 24 | Official leaderboard | |
| FrontierMath Tier 4 v2 | 46.3% | #24 of 68 | Official leaderboard | |
| MysteryMechanism | 35.1% | #11 of 31 | Official leaderboard | |
| ARC-AGI-2 | 61.3% | #35 of 98 | Official leaderboard | |
| ARC-AGI | 89% | #38 of 95 | Official leaderboard | |
| SimpleBench | 59.2% | #45 of 106 | Official leaderboard | |
| BioMysteryBench | 70.4% | #14 of 30 | Official leaderboard | |
| FrontierPhysics | 4.20% | #13 of 22 | Official leaderboard | |
| Morphology-to-molecule matching (Anthropic) | 25.4% | #2 of 3 | Lab model card | |
| Terminal-Bench-Science | 20.6% | #10 of 12 | Lab model card | |
| scBench (LatchBio SingleCellBench) | 56.2% | #6 of 7 | Lab model card | |
| SpatialBench (LatchBio) | 67.7% | #6 of 7 | Lab model card | |
| Biomedical image analysis (Anthropic) | 43.5% | #3 of 3 | Lab model card | |
| Medicinal Chemistry (Anthropic) | 58% | #3 of 3 | Lab model card |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| Vibe Code Bench | 90.4% | #4 of 120 | Official leaderboard | |
| APEX-SWE | 55.1% | #7 of 65 | Official leaderboard | |
| DesignArena 3D Design | 1,327 | #17 of 136 | Official leaderboard | |
| MirrorCode | 78.9% | #5 of 39 | Epoch AI run | |
| SWE-bench Multilingual | 83.7% | #10 of 73 | Lab model card | |
| DesignArena Game Dev | 1,316 | #20 of 143 | Official leaderboard | |
| ALE-Bench | 1,996 | #21 of 126 | Official leaderboard | |
| SWE-Bench Pro | 64.8% | #15 of 81 | Lab model card | |
| DesignArena Data Visualization | 1,278 | #23 of 118 | Official leaderboard | |
| DesignArena Code | 1,278 | #34 of 143 | Official leaderboard | |
| SWE-Marathon v1.1 | 50% | #5 of 20 | Official leaderboard | |
| Code Arena | 1,594 | #33 of 128 | Official leaderboard | |
| Brokk Power Ranking | 19.9 | #5 of 19 | Official leaderboard | |
| Code Migration | 44.6% | #21 of 78 | Official leaderboard | |
| DesignArena UI Component | 1,274 | #38 of 139 | Official leaderboard | |
| FrontierCode 2.0 | 28.9% | #10 of 32 | Official leaderboard | |
| DesignArena Website | 1,254 | #52 of 151 | Official leaderboard | |
| CursorBench | 48.4% | #8 of 21 | cursor.com, via Epoch AI | |
| DesignArena SVG | 1,194 | #41 of 103 | Official leaderboard | |
| Vibe Code Bench 1-100 | 20.4% | #11 of 26 | Official leaderboard | |
| CursorBench 4.0 | 48.4% | #9 of 21 | Official leaderboard | |
| FrontierSWE | 43.8% | #11 of 25 | Official leaderboard | |
| Code Migration v1.1 | 52.4% | #14 of 30 | Official leaderboard | |
| OpenDesign Arena | 82.2 | #10 of 21 | Official leaderboard | |
| ProgramBench tests passed | 82% | #5 of 10 | Lab model card | |
| DAXBench v2 | 79.4% | #24 of 47 | Official leaderboard | |
| WeirdML v3 | 11.7% | #14 of 27 | Official leaderboard | |
| SWE Atlas v1.1 | 59.4% | #16 of 28 | Official leaderboard | |
| Android Bench 2.0 | 20.7% | #10 of 17 | Official leaderboard | |
| SWE-bench Verified Extended | 53.3% | #26 of 44 | Official leaderboard | |
| IOI v2 (Vals) | 47.3% | #30 of 45 | Official leaderboard | |
| FrontierCode 1.1 | 46.4% | #9 of 13 | Lab model card | |
| DesignArena Image to HTML | 1,151 | #44 of 51 | Official leaderboard | |
| SWE-bench Multimodal | 30.7% | #11 of 12 | Lab model card | |
| SWE-Bench Pro V2 Hard | 52.9% | #17 of 18 | Official leaderboard | |
| SWE-Bench Pro V2 | 87.2% | #18 of 18 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| LisanBench | 12897.0 | #11 of 133 | Official leaderboard | |
| LLM Chess | 1,143 | #18 of 123 | Official leaderboard | |
| Chess Puzzles | 55% | #16 of 93 | Epoch AI run | |
| Game Arena Nine Men's Morris | 794 | #10 of 39 | Official leaderboard | |
| DuelLab GameBench 2 | 75.1 | #15 of 57 | Official leaderboard | |
| o11y-bench | 87.5% | #6 of 22 | Official leaderboard | |
| Terminal-Bench 4.0 | 40% | #13 of 40 | Official leaderboard | |
| Terminal-Bench 4.1 | 40% | #14 of 43 | Official leaderboard | |
| Terminal-Bench 2.1 Extended | 37.7% | #15 of 44 | Official leaderboard | |
| Game Arena Dots and Boxes | 958 | #14 of 40 | Official leaderboard | |
| Game Arena Go | 1,076 | #15 of 39 | Official leaderboard | |
| EBR-bench | 43.8% | #12 of 31 | Epoch AI run | |
| Mystery Game Puzzles | 30% | #34 of 84 | Epoch AI run | |
| Game Arena Reversi | 690 | #16 of 39 | Official leaderboard | |
| Game Arena Four in a Row | 419 | #19 of 43 | Official leaderboard | |
| BrowseComp Extended | 27.5% | #19 of 42 | Official leaderboard | |
| Collaborative Intelligence Benchmark | 51.8% | #9 of 18 | Official leaderboard | |
| Agent Arena | 0.47% | #29 of 56 | Official leaderboard | |
| ARC-AGI-3 | 2.82% | #14 of 27 | Official leaderboard | |
| Hermes Bench | 71% | #9 of 17 | Official leaderboard | |
| RuneBench | 4.45 | #46 of 76 | Official leaderboard | |
| GAIA | 37.2% | #39 of 64 | Official leaderboard | |
| Game Arena Dark Hex | 241 | #25 of 41 | Official leaderboard | |
| AutomationBench | 12% | #26 of 42 | Official leaderboard | |
| DRACO | 81.5% | #8 of 12 | Lab model card | |
| Kaggle Game Arena | 257 | #29 of 42 | Official leaderboard | |
| Game Arena Bargaining | 163 | #27 of 39 | Official leaderboard | |
| Game Arena Checkers | 642 | #31 of 41 | Official leaderboard | |
| Game Arena Chess | 594 | #33 of 42 | Official leaderboard | |
| Game Arena Coin Game | 239 | #31 of 39 | Official leaderboard | |
| BU Bench V2.1 | 69.5 | #13 of 16 | Official leaderboard | |
| Game Arena Clobber | 133 | #32 of 39 | Official leaderboard | |
| Game Arena Ultimate Tic-Tac-Toe | 86 | #32 of 39 | Official leaderboard | |
| StarSkirmish Bench | 21% | #8 of 9 | Official leaderboard | |
| Game Arena Word Art | 102 | #36 of 39 | Official leaderboard | |
| Game Arena Word Association | 25 | #37 of 40 | Official leaderboard | |
| Game Arena Lines of Action | 45 | #38 of 39 | Official leaderboard | |
| Cresta-Bench | 20.1% | #14 of 14 | Official leaderboard | |
| WANDR | 49.9% | #4 of 4 | Lab model card | |
| OSWorld 2.1 (offline subset) | 72.4% | #3 of 3 | Lab model card |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MedScribe | 89.5% | #6 of 97 | Official leaderboard | |
| MedCode | 51.5% | #15 of 96 | Official leaderboard | |
| GDPval Extended | 3.10% | #8 of 43 | Official leaderboard | |
| EEBench | 67.1% | #7 of 30 | Official leaderboard | |
| HealthBench | 61.6% | #8 of 31 | Lab model card | |
| HealthBench Professional | 64.8% | #7 of 26 | Lab model card | |
| Vals Finance Agent v2.1 | 65.4% | #11 of 40 | Official leaderboard | |
| Legal Research Bench | 43.8% | #29 of 93 | Official leaderboard | |
| APEX-Accounting | 62.2% | #16 of 49 | Official leaderboard | |
| Tax Agent Bench | 31.9% | #28 of 84 | Official leaderboard | |
| Vals Finance Agent v2 | 54.1% | #28 of 80 | Official leaderboard | |
| Vals Index v2.1 | 54.4% | #19 of 49 | Official leaderboard | |
| DeepFinance | 40% | #9 of 23 | Official leaderboard | |
| APEX-Agents | 56.1% | #24 of 61 | Collected by Epoch AI, source not named | |
| APEX-Agents 1.1 | 68.3% | #28 of 65 | Official leaderboard | |
| YC-Bench | $1,170,559 | #27 of 59 | Official leaderboard | |
| Vals Legal Index | 70.3% | #18 of 38 | Official leaderboard | |
| Vals Index v2.2 | 58.7% | #18 of 37 | Official leaderboard | |
| Excel Modeling Benchmark | 59% | #37 of 76 | Official leaderboard | |
| APEX-1 | 64.7% | #35 of 71 | Official leaderboard | |
| Public Benefits Bench | 63.2% | #25 of 50 | Official leaderboard | |
| Vals Finance Index | 57.5% | #20 of 40 | Official leaderboard | |
| RCM-Bench-1 | 88.4% | #10 of 20 | Official leaderboard | |
| OfficeQA | 73.5% | #8 of 14 | Lab model card | |
| GDPx | 4.94% | #13 of 20 | Official leaderboard | |
| Vending-Bench 2 | $3,602 | #46 of 67 | Official leaderboard | |
| TaxCalcBench v2 | 24% | #23 of 32 | Official leaderboard | |
| Harvey LAB (Legal Agent Benchmark) | 1.25% | #58 of 80 | Official leaderboard | |
| BenchCAD | 67% | #8 of 11 | Lab model card | |
| AA-Briefcase v1.1 | 1,578 | #6 of 8 | Lab model card | |
| PolicyBench | 79.2% | #40 of 50 | Official leaderboard | |
| GDPval-AA v2.1 | 1,620 | #8 of 10 | Lab model card | |
| Excel Modeling Benchmark v1.1 | 65.5% | #34 of 42 | Official leaderboard | |
| OfficeQA Pro | 60.3% | #13 of 16 | Lab model card | |
| MedXpertQA MM Extended | 48.4% | #34 of 41 | Official leaderboard | |
| Immigration Bench | 61.8% | #22 of 25 | Official leaderboard | |
| Harvey LAB v1.1 (HLAB-Vals-1.1) | 4.17% | #34 of 38 | Official leaderboard | |
| BenchCAD 2.0 | 18% | #5 of 5 | Official leaderboard | |
| BenchCAD 2.1 | 22.9% | #5 of 5 | Official leaderboard | |
| PhysicianBench | 43% | #3 of 3 | Lab model card |
Vision, audio and robotics
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CharXiv Extended | 82.9% | #11 of 38 | Official leaderboard | |
| BVB | 68.4 | #18 of 45 | Official leaderboard | |
| Furniture Assembly Benchmark | 47.5% | #14 of 35 | Official leaderboard | |
| Code2Video-Compass | 5.60 | #6 of 15 | Official leaderboard | |
| MineBench | 1,565 | #33 of 75 | Official leaderboard | |
| Blueprint-Bench 2 | 33% | #17 of 38 | Official leaderboard | |
| VoxelBench | 1,731 | #29 of 56 | Official leaderboard | |
| vCX Benchmark | 85.4% | #23 of 42 | Official leaderboard | |
| MMMU-Pro Extended | 31.9% | #23 of 38 | Official leaderboard | |
| Code2Video Arena | 1,479 | #8 of 13 | Official leaderboard | |
| 1H-VideoQA | 68.3% | #13 of 20 | Official leaderboard | |
| ParseBench | 56% | #30 of 44 | Official leaderboard | |
| SplatBench | 3.60 | #6 of 8 | Official leaderboard | |
| Chartography | 46.4% | #11 of 14 | Lab model card | |
| ShadowBench | 41.3 | #5 of 6 | Official leaderboard |
Safety and security
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| BioTIER Refuse | 97.9% | #3 of 69 | Official leaderboard | |
| FORTRESS | 9.48% | #4 of 73 | Official leaderboard | |
| CyberBench v1.1 (Vals) | 75.5% | #6 of 44 | Official leaderboard | |
| Prompt injection robustness against adaptive attackers, browser use (Anthropic) | 0% | #2 of 13 | Lab model card | |
| MASK | 96.8% | #4 of 19 | Lab model card | |
| Prompt injection robustness against adaptive attackers, coding (Anthropic) | 0.08% | #4 of 14 | Lab model card | |
| Prompt injection robustness against adaptive attackers, computer use (Anthropic) | 0.07% | #5 of 14 | Lab model card | |
| Suicide Memory Bench | 1.03 | #5 of 12 | Official leaderboard | |
| SRE-Bench | 5.72% | #15 of 35 | Official leaderboard | |
| Single-turn harmful request evaluations (Anthropic) | 98.4% | #9 of 19 | Lab model card | |
| DistressBench | 71.7% | #20 of 42 | Official leaderboard | |
| Anthropic political even-handedness eval | 97.7% | #7 of 14 | Lab model card | |
| LinuxArena | 6.00% | #3 of 6 | Lab model card | |
| nocot-bench (NCRI) | 68.3 | #108 of 185 | Official leaderboard | |
| Single-turn benign request evaluations (Anthropic) | 0.17% | #12 of 19 | Lab model card | |
| CyScenarioBench | 3.30% | #11 of 17 | Lab model card | |
| Malicious use of Claude Code (Anthropic) | 84.3% | #11 of 17 | Lab model card | |
| ExploitBench | 49% | #14 of 21 | Lab model card | |
| Binary Exploitation Benchmark (Anthropic, OSS-Fuzz based) | 58.2% | #4 of 6 | Lab model card | |
| Virology Capabilities Test (VCT) | 48% | #22 of 31 | Lab model card | |
| Gray Swan IPI Arena benchmark | 0.70% | #8 of 11 | Lab model card | |
| BioSimBench | 0.13 | #9 of 12 | Official leaderboard | |
| Self-preference bias (Anthropic) | 0.23 | #3 of 4 | Lab model card | |
| Malicious computer use (Anthropic) | 82.6% | #11 of 14 | Lab model card | |
| SHADE-Arena | 1.20% | #15 of 16 | Lab model card | |
| BioTIER Permit | 67.8% | #68 of 69 | Official leaderboard | |
| BBQ (Bias Benchmark for QA) | 55.6% | #25 of 25 | Lab model card | |
| ExploitGym | 9.40% | #15 of 15 | Lab model card | |
| Anthropic automated behavioral audit | 3.19 | #6 of 6 | Lab model card | |
| Anthropic undermining liberal democracy (behavioral audit metric) | 1.22 | #6 of 6 | Lab model card | |
| Concealment of changes (Anthropic) | 95.9% | #3 of 3 | Lab model card | |
| Sandbox escape evaluation (Anthropic) | 8.70% | #3 of 3 | Lab model card | |
| Silent copying of answers (Anthropic) | 17.3% | #3 of 3 | Lab model card | |
| Opportunistic sandbox escape (Anthropic) | 4.00% | #2 of 2 | Lab model card |
Other
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| ACCoRD | 81.5 | #9 of 152 | Official leaderboard | |
| SAGE v1.1 (Vals) | 89.8% | #3 of 29 | Official leaderboard | |
| Conceptual Reasoning Index | 65.3 | #23 of 151 | Official leaderboard | |
| LMCA | 51.1% | #27 of 147 | Official leaderboard | |
| LiveBench | 68.9% | #12 of 31 | Official leaderboard | |
| SAGE (Vals) | 45.8% | #38 of 81 | Official leaderboard | |
| Anthropic ECI (AECI) | 167.1 | #3 of 6 | Lab model card | |
| Vals Science Index | 38.5% | #13 of 25 | Official leaderboard | |
| Hermes Index | 37.0 | #10 of 17 | Official leaderboard | |
| Political Science LLM Benchmark | 65.7% | #28 of 42 | Official leaderboard |
Its lab's numbers put it at 156.8, 3.6 points below its ECI
The Self-Reported Index fits one score to the 18 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 154.2 to 159.4. How we compute it.
- Global MMLU 87.8
- MILU 87.6
- SWE-bench Multilingual 83.7
- ProgramBench tests passed 82
- DRACO 81.5
- OfficeQA 73.5
- BenchCAD 67
- SWE-Bench Pro 64.8
- HealthBench Professional 64.8
- HealthBench 61.6
- OfficeQA Pro 60.3
- Humanity's Last Exam 57.4
- FrontierCode 1.1 46.4
- Chartography 46.4
- FrontierSWE 43.8
- Terminal-Bench 4.0 39.2
- SWE-bench Multimodal 30.7
- Terminal-Bench-Science 20.6
Its lab's numbers next to 3 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 |
|---|---|---|---|
| Humanity's Last Exam | 57.4% Multidisciplinary reasoning Humanity's Last Exam (with tools) | 30.1% Leaderboard | +27 pts |
| Terminal-Bench 4.0 | 39.2% Agentic coding Terminal-Bench 4.0 | 40% Leaderboard, max, Claude Code | −0.8 pts |
| FrontierSWE | 43.8% FrontierSWE v2, max effort, Proximal harness, five trials per task (8.6) | 43.8% Leaderboard, mean@5, proximus | 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. 58 of them aren't in the launch post.
System Card: Claude Haiku 5.5 Oct 7, 2026
Capabilities
- SWE-Bench Pro 64.8
- SWE-bench Multilingual 83.7
- SWE-bench Multimodal 30.7
- FrontierCode 1.1 46.4
- Humanity's Last Exam 45.9
- OSWorld 2.1 (offline subset) 72.4
- GDPval-AA v2.1 1620
- AA-Briefcase v1.1 1578
- Terminal-Bench 4.0 39.2
- Terminal-Bench-Science 20.6
- FrontierSWE 43.8
- ProgramBench tests passed 82
- DRACO 81.5
- WANDR 49.9
- Chartography 46.4
- BenchCAD 67
- OfficeQA 73.5
- OfficeQA Pro 60.3
- Global MMLU 87.8
- MILU 87.6
- Morphology-to-molecule matching (Anthropic) 25.4
- Medicinal Chemistry (Anthropic) 58
- Protein Design (Anthropic)
- Biomedical image analysis (Anthropic) 43.5
Health
- HealthBench Professional 64.8
- HealthBench 61.6
- PhysicianBench 43
Hallucination
Safety
- Single-turn harmful request evaluations (Anthropic) 98.4
- Single-turn benign request evaluations (Anthropic) 0.17
- 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) 84.3
- Malicious computer use (Anthropic) 82.6
Robustness
Alignment
- Anthropic automated behavioral audit 3.19
- Anthropic undermining liberal democracy (behavioral audit metric) 1.22
- Sandbox escape evaluation (Anthropic) 8.7
- Opportunistic sandbox escape (Anthropic) 4
- Self-preference bias (Anthropic) 0.23
- MASK 96.8
- Silent copying of answers (Anthropic) 17.3
- Concealment of changes (Anthropic) 95.9
Monitorability
Bio and chem risk
Cyber risk
AI R&D risk
- Anthropic ECI (AECI) 167.1