Knowledge & reasoning
HealthBench (length-adjusted) leaderboard
As of 2026-10-07, the highest measured score on HealthBench (length-adjusted) is 65.4% by Claude Sonnet 5.5. 196 more models have estimated scores, calibrated from the benchmarks they were measured on.
Measured scores: benchlm.ai.
| # | Model | Score | Source |
|---|---|---|---|
| 1 | Claude Sonnet 5.5 | 65.4% | measured |
| 2 | Claude Opus 5.5 | 60.6% | measured |
| 3 | Ling 3.1 Flash | 60.6% | estimated ± 2.2 pp, medium confidence |
| 4 | Gemini 4 Argon | 59.5% | estimated ± 3.7 pp, medium confidence |
| 5 | GPT-6.1 Sol | 58.5% | measured |
| 6 | Claude Fable 5 | 58.3% | estimated ± 2.5 pp, low confidence |
| 7 | GPT-6 Astra | 58.3% | measured |
| 8 | Claude Fable 5.1 | 58.0% | estimated ± 2.5 pp, low confidence |
| 9 | Claude Opus 5 | 57.8% | measured |
| 10 | Gemini 3.7 Flash | 57.2% | estimated ± 2.5 pp, low confidence |
| 11 | GPT-5.5 Pro | 57.0% | estimated ± 2.5 pp, low confidence |
| 12 | GPT-5.4 Pro | 56.8% | estimated ± 2.5 pp, low confidence |
| 13 | Kimi K3 | 56.8% | estimated ± 2.5 pp, low confidence |
| 14 | Claude Opus 4.6 (Adaptive) | 56.3% | estimated ± 2.5 pp, low confidence |
| 15 | Gemini 3.5 Flash | 56.1% | estimated ± 2.5 pp, low confidence |
| 16 | Gemini 3.6 Flash | 55.6% | estimated ± 2.5 pp, low confidence |
| 17 | DeepSeek V4 Pro 0813 | 55.1% | estimated ± 2.5 pp, low confidence |
| 18 | DeepSeek V4 Flash 0731 | 54.7% | estimated ± 2.5 pp, low confidence |
| 19 | GPT-6 Luna | 54.5% | measured |
| 20 | Muse Spark 1.3 | 54.4% | estimated ± 3.7 pp, medium confidence |
| 21 | MiMo-V2.6-Pro | 54.1% | estimated ± 3.7 pp, medium confidence |
| 22 | Qwen3.8 Max Preview | 54.0% | estimated ± 3.7 pp, medium confidence |
| 23 | GLM-5.3 | 54.0% | estimated ± 3.7 pp, medium confidence |
| 24 | Step 5 Preview | 53.9% | estimated ± 3.7 pp, medium confidence |
| 25 | Claude Haiku 5.5 | 53.9% | estimated ± 3.7 pp, medium confidence |
| 26 | GLM-5.3-Flash | 53.9% | estimated ± 3.7 pp, medium confidence |
| 27 | Hy3 Preview | 53.9% | estimated ± 3.7 pp, medium confidence |
| 28 | Qwen3.8-Flash-Next | 53.9% | estimated ± 3.7 pp, medium confidence |
| 29 | Muse Spark 1.2 | 53.9% | estimated ± 3.7 pp, medium confidence |
| 30 | DeepSeek V4.1 Flash | 53.9% | estimated ± 3.7 pp, medium confidence |
| 31 | Mistral Large 4 | 53.9% | estimated ± 3.7 pp, medium confidence |
| 32 | Claude Sonnet 5 | 53.9% | estimated ± 3.7 pp, medium confidence |
| 33 | Grok 4.3 | 53.9% | estimated ± 3.7 pp, low confidence |
| 34 | MiMo-V2.6-Flash | 53.9% | estimated ± 3.7 pp, low confidence |
| 35 | GLM-5.2 | 53.9% | estimated ± 3.7 pp, low confidence |
| 36 | GPT-5.3 Codex | 53.9% | estimated ± 3.7 pp, low confidence |
| 37 | Qwen3.8-27B | 53.9% | estimated ± 3.7 pp, low confidence |
| 38 | A.X K2 | 53.9% | estimated ± 3.7 pp, low confidence |
| 39 | Apodex 1.1 | 53.9% | estimated ± 3.7 pp, low confidence |
| 40 | Apodex 1.1 Mini | 53.9% | estimated ± 3.7 pp, low confidence |
| 41 | Celeris-1 | 53.9% | estimated ± 3.7 pp, low confidence |
| 42 | Claude 3 Haiku | 53.9% | estimated ± 3.7 pp, low confidence |
| 43 | Claude 3 Opus | 53.9% | estimated ± 3.7 pp, low confidence |
| 44 | Claude 4.1 Opus | 53.9% | estimated ± 3.7 pp, low confidence |
| 45 | Claude 4.1 Opus Thinking | 53.9% | estimated ± 3.7 pp, low confidence |
| 46 | Claude 4 Sonnet | 53.9% | estimated ± 3.7 pp, low confidence |
| 47 | Claude Opus 4.5 | 53.9% | estimated ± 3.7 pp, low confidence |
| 48 | Claude Opus 4.6 | 53.9% | estimated ± 3.7 pp, low confidence |
| 49 | Claude Opus 4.7 | 53.9% | estimated ± 3.7 pp, low confidence |
| 50 | Command A+ | 53.9% | estimated ± 3.7 pp, low confidence |
| 51 | DeepSeek-R1 | 53.9% | estimated ± 3.7 pp, low confidence |
| 52 | DeepSeek R1 Distill Qwen 32B | 53.9% | estimated ± 3.7 pp, low confidence |
| 53 | DeepSeek V3 | 53.9% | estimated ± 3.7 pp, low confidence |
| 54 | DeepSeek V3 0324 | 53.9% | estimated ± 3.7 pp, low confidence |
| 55 | DeepSeek V3.1 | 53.9% | estimated ± 3.7 pp, low confidence |
| 56 | DeepSeek V3.1 (Reasoning) | 53.9% | estimated ± 3.7 pp, low confidence |
| 57 | DeepSeek V3.2 | 53.9% | estimated ± 3.7 pp, low confidence |
| 58 | Exaone 4.0 1.2B | 53.9% | estimated ± 3.7 pp, low confidence |
| 59 | Exaone 4.0 32B | 53.9% | estimated ± 3.7 pp, low confidence |
| 60 | Gemini 1.0 Pro | 53.9% | estimated ± 3.7 pp, low confidence |
| 61 | Gemini 1.5 Pro | 53.9% | estimated ± 3.7 pp, low confidence |
| 62 | Gemini 2.5 Flash | 53.9% | estimated ± 3.7 pp, low confidence |
| 63 | Gemini 2.5 Pro | 53.9% | estimated ± 3.7 pp, low confidence |
| 64 | Gemini 3.5 Flash-Lite | 53.9% | estimated ± 3.7 pp, low confidence |
| 65 | Gemini 3 Flash | 53.9% | estimated ± 3.7 pp, low confidence |
| 66 | Gemma 3 27B | 53.9% | estimated ± 3.7 pp, low confidence |
| 67 | Gemma 4 12B | 53.9% | estimated ± 3.7 pp, low confidence |
| 68 | Gemma 4 26B A4B | 53.9% | estimated ± 3.7 pp, low confidence |
| 69 | Gemma 4 31B | 53.9% | estimated ± 3.7 pp, low confidence |
| 70 | Gemma 4 E2B | 53.9% | estimated ± 3.7 pp, low confidence |
| 71 | Gemma 4 E4B | 53.9% | estimated ± 3.7 pp, low confidence |
| 72 | GLM-4.5-Air | 53.9% | estimated ± 3.7 pp, low confidence |
| 73 | GLM-4.6 | 53.9% | estimated ± 3.7 pp, low confidence |
| 74 | GLM-4.7 | 53.9% | estimated ± 3.7 pp, low confidence |
| 75 | GLM-5 | 53.9% | estimated ± 3.7 pp, low confidence |
| 76 | GLM-5.1 | 53.9% | estimated ± 3.7 pp, low confidence |
| 77 | GLM-5-Turbo | 53.9% | estimated ± 3.7 pp, low confidence |
| 78 | GLM-5V-Turbo | 53.9% | estimated ± 3.7 pp, low confidence |
| 79 | GPT-4.1 | 53.9% | estimated ± 3.7 pp, low confidence |
| 80 | GPT-4.1 mini | 53.9% | estimated ± 3.7 pp, low confidence |
| 81 | GPT-4.1 nano | 53.9% | estimated ± 3.7 pp, low confidence |
| 82 | GPT-4 Turbo | 53.9% | estimated ± 3.7 pp, low confidence |
| 83 | GPT-4o | 53.9% | estimated ± 3.7 pp, low confidence |
| 84 | GPT-4o mini | 53.9% | estimated ± 3.7 pp, low confidence |
| 85 | GPT-5.1-Codex | 53.9% | estimated ± 3.7 pp, low confidence |
| 86 | GPT-5.1-Codex-Max | 53.9% | estimated ± 3.7 pp, low confidence |
| 87 | GPT-5.2-Codex | 53.9% | estimated ± 3.7 pp, low confidence |
| 88 | GPT-5 (high) | 53.9% | estimated ± 3.7 pp, low confidence |
| 89 | GPT-5 (medium) | 53.9% | estimated ± 3.7 pp, low confidence |
| 90 | GPT-OSS 120B | 53.9% | estimated ± 3.7 pp, low confidence |
| 91 | GPT-OSS 20B | 53.9% | estimated ± 3.7 pp, low confidence |
| 92 | Granite-4.0-350M | 53.9% | estimated ± 3.7 pp, low confidence |
| 93 | Granite-4.0-H-1B | 53.9% | estimated ± 3.7 pp, low confidence |
| 94 | Granite-4.0-H-350M | 53.9% | estimated ± 3.7 pp, low confidence |
| 95 | Granite 4.2 30B | 53.9% | estimated ± 3.7 pp, low confidence |
| 96 | Granite 4.2 3B | 53.9% | estimated ± 3.7 pp, low confidence |
| 97 | Granite 4.2 8B | 53.9% | estimated ± 3.7 pp, low confidence |
| 98 | Grok 4 | 53.9% | estimated ± 3.7 pp, low confidence |
| 99 | Grok 4.1 Fast | 53.9% | estimated ± 3.7 pp, low confidence |
| 100 | Grok 4.1 Fast (Reasoning) | 53.9% | estimated ± 3.7 pp, low confidence |
| 101 | Grok 4 Fast (Reasoning) | 53.9% | estimated ± 3.7 pp, low confidence |
| 102 | Grok Code Fast 1 | 53.9% | estimated ± 3.7 pp, low confidence |
| 103 | Hy3 | 53.9% | estimated ± 3.7 pp, low confidence |
| 104 | Inkling | 53.9% | estimated ± 3.7 pp, low confidence |
| 105 | K-Exaone | 53.9% | estimated ± 3.7 pp, low confidence |
| 106 | K-EXAONE 2.0 | 53.9% | estimated ± 3.7 pp, low confidence |
| 107 | Kimi K2.6 | 53.9% | estimated ± 3.7 pp, low confidence |
| 108 | Kimi K2 | 53.9% | estimated ± 3.7 pp, low confidence |
| 109 | Kimi K2.5 | 53.9% | estimated ± 3.7 pp, low confidence |
| 110 | Kimi K2.5 (Reasoning) | 53.9% | estimated ± 3.7 pp, low confidence |
| 111 | Kimi K2.7 Code | 53.9% | estimated ± 3.7 pp, low confidence |
| 112 | LFM2.5-2.6B | 53.9% | estimated ± 3.7 pp, low confidence |
| 113 | LFM2.5-8B-A1B | 53.9% | estimated ± 3.7 pp, low confidence |
| 114 | LFM2.5-VL-1.6B-Extract | 53.9% | estimated ± 3.7 pp, low confidence |
| 115 | Ling 2.6 Flash | 53.9% | estimated ± 3.7 pp, low confidence |
| 116 | Ling 3.0 Flash | 53.9% | estimated ± 3.7 pp, low confidence |
| 117 | Ling 3.0 Flash FP8 | 53.9% | estimated ± 3.7 pp, low confidence |
| 118 | Ling 3.0 Flash VL | 53.9% | estimated ± 3.7 pp, low confidence |
| 119 | Ling 3.0 Tiny | 53.9% | estimated ± 3.7 pp, low confidence |
| 120 | Llama 3.1 405B | 53.9% | estimated ± 3.7 pp, low confidence |
| 121 | Llama 4 Maverick | 53.9% | estimated ± 3.7 pp, low confidence |
| 122 | Llama 4 Scout | 53.9% | estimated ± 3.7 pp, low confidence |
| 123 | Mercury 2.5 | 53.9% | estimated ± 3.7 pp, low confidence |
| 124 | MiMo-V2.5-Pro | 53.9% | estimated ± 3.7 pp, low confidence |
| 125 | MiMo-V2-Flash | 53.9% | estimated ± 3.7 pp, low confidence |
| 126 | MiMo-V2-Omni | 53.9% | estimated ± 3.7 pp, low confidence |
| 127 | MiMo-V2-Pro | 53.9% | estimated ± 3.7 pp, low confidence |
| 128 | MiniCPM5-2B | 53.9% | estimated ± 3.7 pp, low confidence |
| 129 | MiniMax M2.7 | 53.9% | estimated ± 3.7 pp, low confidence |
| 130 | MiniMax M3 | 53.9% | estimated ± 3.7 pp, low confidence |
| 131 | Mistral Large 2 | 53.9% | estimated ± 3.7 pp, low confidence |
| 132 | Mistral Large 3 | 53.9% | estimated ± 3.7 pp, low confidence |
| 133 | Mistral Medium 3 | 53.9% | estimated ± 3.7 pp, low confidence |
| 134 | Mistral Medium 3.5 128B | 53.9% | estimated ± 3.7 pp, low confidence |
| 135 | Mistral Small 4 | 53.9% | estimated ± 3.7 pp, low confidence |
| 136 | Mistral Small 4 (Reasoning) | 53.9% | estimated ± 3.7 pp, low confidence |
| 137 | Muse Glimmer 30B | 53.9% | estimated ± 3.7 pp, low confidence |
| 138 | Muse Spark | 53.9% | estimated ± 3.7 pp, low confidence |
| 139 | Nemotron 3.5 Lightning 30B A3B NVFP4 | 53.9% | estimated ± 3.7 pp, low confidence |
| 140 | Nemotron 3 Nano 30B | 53.9% | estimated ± 3.7 pp, low confidence |
| 141 | Nemotron 3 Nano Omni 30B A3B | 53.9% | estimated ± 3.7 pp, low confidence |
| 142 | Nemotron 3 Super 100B | 53.9% | estimated ± 3.7 pp, low confidence |
| 143 | Nemotron 3 Ultra | 53.9% | estimated ± 3.7 pp, low confidence |
| 144 | Nemotron Ultra 253B | 53.9% | estimated ± 3.7 pp, low confidence |
| 145 | North Mini Code | 53.9% | estimated ± 3.7 pp, low confidence |
| 146 | Nova Pro | 53.9% | estimated ± 3.7 pp, low confidence |
| 147 | o1 | 53.9% | estimated ± 3.7 pp, low confidence |
| 148 | o1-preview | 53.9% | estimated ± 3.7 pp, low confidence |
| 149 | o1-pro | 53.9% | estimated ± 3.7 pp, low confidence |
| 150 | o3 | 53.9% | estimated ± 3.7 pp, low confidence |
| 151 | o3-mini | 53.9% | estimated ± 3.7 pp, low confidence |
| 152 | o3-pro | 53.9% | estimated ± 3.7 pp, low confidence |
| 153 | Phi-4 | 53.9% | estimated ± 3.7 pp, low confidence |
| 154 | Phi-4 Multimodal Instruct | 53.9% | estimated ± 3.7 pp, low confidence |
| 155 | Quasar 438B | 53.9% | estimated ± 3.7 pp, low confidence |
| 156 | Qwen2.5 Coder 32B Instruct | 53.9% | estimated ± 3.7 pp, low confidence |
| 157 | Qwen3.5-122B-A10B | 53.9% | estimated ± 3.7 pp, low confidence |
| 158 | Qwen3.5-27B | 53.9% | estimated ± 3.7 pp, low confidence |
| 159 | Qwen3.5-35B-A3B | 53.9% | estimated ± 3.7 pp, low confidence |
| 160 | Qwen3.5 397B | 53.9% | estimated ± 3.7 pp, low confidence |
| 161 | Qwen3.5 397B (Reasoning) | 53.9% | estimated ± 3.7 pp, low confidence |
| 162 | Qwen3.6-27B | 53.9% | estimated ± 3.7 pp, low confidence |
| 163 | Qwen3.6-35B-A3B | 53.9% | estimated ± 3.7 pp, low confidence |
| 164 | Qwen 3.6 Max (preview) | 53.9% | estimated ± 3.7 pp, low confidence |
| 165 | Qwen3.6 Plus | 53.9% | estimated ± 3.7 pp, low confidence |
| 166 | Qwen3.7 Max | 53.9% | estimated ± 3.7 pp, low confidence |
| 167 | Qwen3.7 Plus | 53.9% | estimated ± 3.7 pp, low confidence |
| 168 | Qwen3 Max | 53.9% | estimated ± 3.7 pp, low confidence |
| 169 | Qwen3-Omni-30B-A3B-Instruct | 53.9% | estimated ± 3.7 pp, low confidence |
| 170 | Qwen3-Omni-30B-A3B-Thinking | 53.9% | estimated ± 3.7 pp, low confidence |
| 171 | Sarvam 105B | 53.9% | estimated ± 3.7 pp, low confidence |
| 172 | Sarvam 30B | 53.9% | estimated ± 3.7 pp, low confidence |
| 173 | Solar Pro 2 | 53.9% | estimated ± 3.7 pp, low confidence |
| 174 | Solar Pro 3 | 53.9% | estimated ± 3.7 pp, low confidence |
| 175 | Solar Pro 4 | 53.9% | estimated ± 3.7 pp, low confidence |
| 176 | Step 3.7 Flash | 53.9% | estimated ± 3.7 pp, low confidence |
| 177 | Trinity-Large-Preview | 53.9% | estimated ± 3.7 pp, low confidence |
| 178 | Trinity-Large-Thinking | 53.9% | estimated ± 3.7 pp, low confidence |
| 179 | Ultravox v0.6 Llama 3.3 70B | 53.9% | estimated ± 3.7 pp, low confidence |
| 180 | Gemini 3.8 Flash | 53.9% | estimated ± 0.9 pp, medium confidence |
| 181 | GPT-5.6 Sol | 53.9% | estimated ± 0.9 pp, medium confidence |
| 182 | Claude Opus 4.7 (Adaptive) | 53.8% | estimated ± 0.9 pp, medium confidence |
| 183 | Claude Opus 4.8 | 53.8% | estimated ± 0.9 pp, medium confidence |
| 184 | Gemini 3.1 Pro | 53.8% | estimated ± 0.9 pp, medium confidence |
| 185 | GPT-5.4 | 53.8% | estimated ± 0.9 pp, medium confidence |
| 186 | GPT-5.5 | 53.8% | estimated ± 0.9 pp, medium confidence |
| 187 | GPT-5.6 Luna | 53.8% | estimated ± 0.9 pp, medium confidence |
| 188 | GPT-5.6 Terra | 53.8% | estimated ± 0.9 pp, medium confidence |
| 189 | Grok 4.20 | 53.8% | estimated ± 0.9 pp, low confidence |
| 190 | Grok 4.5 | 53.8% | estimated ± 0.9 pp, medium confidence |
| 191 | Grok 4.6 | 53.8% | estimated ± 0.9 pp, medium confidence |
| 192 | GPT-5.2 | 53.6% | estimated ± 2.5 pp, low confidence |
| 193 | Claude Sonnet 4.6 | 53.5% | estimated ± 2.5 pp, low confidence |
| 194 | GPT-6 Sol | 53.2% | measured |
| 195 | Inkling-Small | 52.7% | estimated ± 2.5 pp, low confidence |
| 196 | Muse Spark 1.1 | 52.6% | estimated ± 2.2 pp, low confidence |
| 197 | Claude Opus 4.5 Thinking | 51.1% | estimated ± 2.5 pp, low confidence |
| 198 | Grok 4.7 | 49.4% | estimated ± 2.2 pp, low confidence |
| 199 | Gemini 3 Pro | 48.9% | estimated ± 2.5 pp, low confidence |
| 200 | GPT-5.1 | 47.9% | estimated ± 2.5 pp, low confidence |
| 201 | GPT-5.4 mini | 43.7% | estimated ± 2.5 pp, low confidence |
| 202 | Claude Sonnet 4.5 Thinking | 43.6% | estimated ± 2.5 pp, low confidence |
| 203 | GPT-5.4 nano | 37.4% | estimated ± 2.5 pp, low confidence |
All leaderboards
Agentic · terminal
- AA Terminal-Bench 2.1
- AA Terminal-Bench 4.0
- Terminal-Bench 2.0
- Terminal-Bench 3.0
- Terminal-Bench 4.0
- Terminal-Bench-Science 0.1
- Terminal-Bench 2.1
- Terminal-Bench 2.1 (Vals AI)
Agentic · tools
- AA Agentic Index
- AA-AnalystAgent
- AA AutomationBench
- AA EnterpriseOps-Gym
- GDP.pdf
- AA Harvey LAB
- AA ITBench
- AA Tau3 Banking
- Agents' Last Exam
- APEX-Agents
- APEX-Agents-AA
- ApprenticeBench
- AutomationBench
- BFCL v4
- BrowseComp
- Claw-Eval
- CWE-bench v1
- CyberGym
- DeepPlanning
- DeepSearchQA
- DRACO
- ExploitGym
- GDPval-AA
- Gert Labs
- HLE w/ tools
- JobBench
- MCP Atlas
- MCP-Tasks
- OSWorld-Verified
- OSWorld 2.0
- QwenClawBench
- ResearchClawBench
- SkillsBench
- τ²-bench results
- τ³-bench results
- Toolathlon
- Toolathlon-Verified
- VITA-Bench
- WideResearch
Coding
- AA Coding Index
- AA-SciCode
- CursorBench 3.1
- CursorBench 3.2
- CursorBench 4.0
- DeepSWE
- FrontierCode 1.1 Main
- FrontierCode 1.1 Extended
- FrontierSWE v2
- LiveCodeBench
- LiveCodeBench Pro
- LiveCodeBench v6
- NL2Repo
- OpenHarmony Bench
- PostTrainBench v1.1
- ProgramBench
- React Native Evals
- SciCode
- SWE-bench Pro
- SWE-bench Verified
- SWE-Rebench
- Vals LiveCodeBench
- Vals SWE-bench
- Vibe Code Bench
- VulcanBench v3
Math
- AIME 2025
- AIME26
- FrontierMath (legacy)
- FrontierMath v2 (Tier 4)
- FrontierMath v2 (Tiers 1-3)
- HMMT Feb 2025
- HMMT Feb 2026
- HMMT Nov 2025
- IMOAnswerBench
- MATH-500
- MMAnswerBench
Knowledge & reasoning
- AA-GPQA Diamond
- AA-HLE
- ARC-AGI-2
- ARC-AGI-1
- ARC-AGI-3
- Artificial Analysis Intelligence Index
- BioMysteryBench (human-difficult)
- BioMysteryBench (human-solvable)
- C-Eval
- CritPt
- GPQA
- GPQA Diamond
- HealthBench (raw)
- HealthBench Hard
- HealthBench (length-adjusted)
- HealthBench Professional
- HealthBench Professional (raw)
- HLE
- HLE w/o tools
- HLE-Verified
- JevBench 1.4
- LABBench2
- MedXpertQA (Text)
- MMLU
- MMLU-Pro
- MMLU-Pro (Arcee)
- MMLU-Redux
- MMMLU
- AA-Omniscience Accuracy
- SuperGPQA
- Vals GPQA Diamond
- Vals MMLU-Pro
Instruction following
Multilingual
Vision & documents
- AA-MMMU-Pro
- BabyVision
- CharXiv
- CharXiv w/o tools
- ERQA
- MathVision
- MathVision w/ Python
- MedXpertQA (MM)
- MMMU
- MMMU-Pro
- MMMU-Pro w/ Python
- MMVU
- OfficeQA Pro
- OmniDocBench 1.5
- RealWorldQA
- RefCOCO (avg)
- ScreenSpot Pro
- SimpleVQA
- Video-MME (with subtitle)
- VideoMMMU
- V*
- ZeroBench