benchgap
Knowledge & reasoning

HealthBench (raw) leaderboard

As of 2026-11-21, the highest measured score on HealthBench (raw) is 69.4% by Claude Sonnet 5.5. 120 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Fable 5.274.8%estimated ± 6.0 pp, low confidence
2Claude Sonnet 5.569.4%measured
3Claude Opus 5.568.1%measured
4Claude Opus 567.1%measured
5Claude Fable 5.164.5%estimated ± 6.0 pp, medium confidence
6Gemini 4 Carbon64.1%estimated ± 7.6 pp, low confidence
7Gemini 4 Argon63.0%estimated ± 6.0 pp, medium confidence
8Claude Fable 561.8%estimated ± 6.0 pp, medium confidence
9Muse Spark60.3%estimated ± 0.7 pp, low confidence
10Muse Wonder Full60.3%estimated ± 6.0 pp, medium confidence
11Kimi K3.1 Pro59.4%estimated ± 6.0 pp, medium confidence
12GPT-5.459.3%estimated ± 0.7 pp, low confidence
13Muse Wonder57.5%estimated ± 6.0 pp, medium confidence
14GPT-6 Astra56.9%measured
15MiMo-V2.6-Pro56.9%estimated ± 6.0 pp, medium confidence
16GPT-6.1 Sol56.7%measured
17Claude Opus 4.856.3%estimated ± 6.0 pp, medium confidence
18Muse Spark 1.356.3%estimated ± 6.0 pp, medium confidence
19MiniMax M3.1 Flash55.7%estimated ± 6.0 pp, medium confidence
20Gemini 3.7 Flash55.6%estimated ± 6.0 pp, medium confidence
21Gemini 3.8 Flash55.6%estimated ± 6.0 pp, medium confidence
22Gemini 3.1 Pro54.9%estimated ± 6.0 pp, medium confidence
23Kimi K354.8%estimated ± 6.0 pp, medium confidence
24GPT-6.2 Astra54.7%estimated ± 1.6 pp, high confidence
25Step 5 Preview54.5%estimated ± 6.0 pp, medium confidence
26Muse Spark 1.154.2%estimated ± 6.0 pp, medium confidence
27GPT-6.2 Sol54.1%estimated ± 1.6 pp, high confidence
28GPT-5.553.9%estimated ± 6.0 pp, medium confidence
29Muse Spark 1.253.6%estimated ± 6.0 pp, medium confidence
30GPT-5.6 Sol53.1%estimated ± 0.7 pp, medium confidence
31Claude Haiku 5.552.7%estimated ± 6.0 pp, medium confidence
32GPT-5.6 Terra52.5%estimated ± 0.7 pp, medium confidence
33Inkling 2 Small52.4%estimated ± 6.0 pp, medium confidence
34Muse Spark Open52.2%estimated ± 6.0 pp, medium confidence
35Grok 4.751.5%estimated ± 6.0 pp, medium confidence
36Qwen3.8 Max Preview51.5%estimated ± 6.0 pp, medium confidence
37GPT-5.6 Luna51.3%estimated ± 0.7 pp, medium confidence
38Grok 4.651.3%estimated ± 6.0 pp, medium confidence
39Gemini 3.5 Flash51.2%estimated ± 6.0 pp, medium confidence
40Grok 4.551.2%estimated ± 6.0 pp, medium confidence
41GPT-6.1 Terra51.1%measured
42GPT-5.3 Codex51.0%estimated ± 6.0 pp, medium confidence
43Claude Opus 4.750.8%estimated ± 6.0 pp, medium confidence
44GLM-5.350.8%estimated ± 6.0 pp, medium confidence
45GPT-6 Luna50.0%measured
46Claude Sonnet 549.9%estimated ± 6.0 pp, medium confidence
47GLM-5.249.7%estimated ± 6.0 pp, medium confidence
48DeepSeek V4 Pro49.6%estimated ± 6.0 pp, medium confidence
49Gemini 3.6 Flash49.4%estimated ± 6.0 pp, medium confidence
50Qwen3.7 Max49.2%estimated ± 6.0 pp, medium confidence
51GPT-5.5 Pro48.7%estimated ± 7.6 pp, low confidence
52Claude Opus 4.6 (Adaptive)48.6%estimated ± 6.0 pp, medium confidence
53Gemini 3 Pro48.4%estimated ± 6.0 pp, medium confidence
54GPT-5.4 Pro48.4%estimated ± 7.6 pp, low confidence
55Ling 3.1 Flash48.2%estimated ± 6.0 pp, medium confidence
56Claude Sonnet 4.648.1%estimated ± 7.6 pp, low confidence
57DeepSeek V4.1 Flash 101748.0%estimated ± 6.0 pp, medium confidence
58MiniMax M347.8%estimated ± 6.0 pp, medium confidence
59DeepSeek V4 Flash 073147.4%estimated ± 6.0 pp, medium confidence
60GPT-6 Sol47.1%measured
61Qwen3.8-Flash-Next46.8%estimated ± 6.0 pp, low confidence
62GPT-5.246.6%estimated ± 6.0 pp, low confidence
63Kimi K2.646.4%estimated ± 6.0 pp, low confidence
64Grok 4.346.1%estimated ± 6.0 pp, low confidence
65GPT-5.2-Codex44.6%estimated ± 6.0 pp, low confidence
66MiMo-V2.5-Pro44.6%estimated ± 6.0 pp, low confidence
67Qwen3.7 Plus44.5%estimated ± 6.0 pp, low confidence
68MiMo-V2.6-Flash44.1%estimated ± 6.0 pp, low confidence
69Kimi K2.7 Code44.0%estimated ± 6.0 pp, low confidence
70Beam 1.0 501B-A23B43.7%estimated ± 6.0 pp, low confidence
71Apodex 1.143.1%estimated ± 6.0 pp, low confidence
72Apodex 1.1 Mini43.1%estimated ± 6.0 pp, low confidence
73Qwen3.8-27B42.9%estimated ± 6.0 pp, low confidence
74Hy342.5%estimated ± 6.0 pp, low confidence
75Hy3 Preview42.5%estimated ± 6.0 pp, low confidence
76Inkling-Small42.3%estimated ± 6.0 pp, low confidence
77Inkling40.9%estimated ± 6.0 pp, low confidence
78Qwen 3.6 Max (preview)39.7%estimated ± 6.0 pp, low confidence
79Kimi K2.539.6%estimated ± 6.0 pp, low confidence
80Nemotron 3.5 Super VL 120B A12B39.6%estimated ± 6.0 pp, low confidence
81Nemotron 3.5 Super VL 120B A12B NVFP439.6%estimated ± 6.0 pp, low confidence
82MiMo-V2-Pro39.3%estimated ± 6.0 pp, low confidence
83Claude Opus 4.5 Thinking39.0%estimated ± 6.0 pp, low confidence
84GLM-5.139.0%estimated ± 6.0 pp, low confidence
85Solar Pro 438.7%estimated ± 6.0 pp, low confidence
86A.X K238.5%estimated ± 6.0 pp, low confidence
87MiniMax M2.738.5%estimated ± 6.0 pp, low confidence
88GLM-538.2%estimated ± 6.0 pp, low confidence
89Qwen3.5 397B37.8%estimated ± 6.0 pp, low confidence
90GPT-537.3%estimated ± 6.0 pp, low confidence
91GPT-5.137.3%estimated ± 6.0 pp, low confidence
92Nemotron 3 Ultra37.2%estimated ± 6.0 pp, low confidence
93GPT-5.4 nano37.1%estimated ± 6.0 pp, low confidence
94GPT-5.4 mini36.9%estimated ± 6.0 pp, low confidence
95GLM-5-Turbo36.6%estimated ± 6.0 pp, low confidence
96Qwen3.6 Plus36.6%estimated ± 6.0 pp, low confidence
97GLM-4.736.1%estimated ± 6.0 pp, low confidence
98Grok 435.4%estimated ± 6.0 pp, low confidence
99GPT-5.1-Codex34.3%estimated ± 6.0 pp, low confidence
100GPT-5.1-Codex-Max34.3%estimated ± 6.0 pp, low confidence
101Qwen3.5-122B-A10B33.7%estimated ± 6.0 pp, low confidence
102Nemotron 3.5 Bolt 74B-A9B33.2%estimated ± 6.0 pp, low confidence
103Qwen3.5-27B32.3%estimated ± 6.0 pp, low confidence
104Ling 3.0 Flash32.0%estimated ± 6.0 pp, low confidence
105Gemma 4 31B31.9%estimated ± 6.0 pp, low confidence
106Qwen3.6-27B31.4%estimated ± 6.0 pp, low confidence
107Gemini 2.5 Pro30.7%estimated ± 6.0 pp, low confidence
108Qwen3.6-35B-A3B30.3%estimated ± 6.0 pp, low confidence
109MiMo-V2-Omni30.2%estimated ± 6.0 pp, low confidence
110Ling 3.0 Flash VL30.1%estimated ± 6.0 pp, low confidence
111Muse Glimmer 30B30.1%estimated ± 6.0 pp, low confidence
112Step 3.7 Flash29.4%estimated ± 6.0 pp, low confidence
113Qwen3.5-35B-A3B28.9%estimated ± 6.0 pp, low confidence
114o327.8%estimated ± 6.0 pp, low confidence
115GPT-OSS 120B27.3%estimated ± 6.0 pp, low confidence
116Gemma 4 26B A4B26.9%estimated ± 6.0 pp, low confidence
117Grok 4.1 Fast (Reasoning)26.9%estimated ± 6.0 pp, low confidence
118Claude Opus 4.626.6%estimated ± 6.0 pp, low confidence
119Grok 4 Fast (Reasoning)26.6%estimated ± 6.0 pp, low confidence
120Gemini 3.5 Flash-Lite26.3%estimated ± 6.0 pp, low confidence
121Quasar 438B26.2%estimated ± 6.0 pp, low confidence
122K-EXAONE 2.026.0%estimated ± 6.0 pp, low confidence
123GLM-5V-Turbo24.2%estimated ± 6.0 pp, low confidence
124o4-mini23.4%estimated ± 6.0 pp, low confidence
125DeepSeek-R122.6%estimated ± 6.0 pp, low confidence
126Trinity-Large-Preview22.6%estimated ± 6.0 pp, low confidence
127Trinity-Large-Thinking22.6%estimated ± 6.0 pp, low confidence
128Gemma 4 12B22.4%estimated ± 6.0 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General