benchgap
Knowledge & reasoning

HealthBench Hard leaderboard

As of 2026-11-21, the highest measured score on HealthBench Hard is 42.8% by Muse Spark. 206 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Muse Spark42.8%measured
2GPT-5.440.1%measured
3Claude Opus 5.539.7%estimated ± 0.6 pp, low confidence
4Sakana Fugu36.6%estimated ± 5.1 pp, low confidence
5Sakana Fugu-Ultra36.6%estimated ± 5.1 pp, low confidence
6GPT-6 Astra36.6%measured
7Claude Opus 4.736.5%estimated ± 5.1 pp, low confidence
8Kimi K3.1 Pro36.5%estimated ± 5.1 pp, low confidence
9Claude Mythos 536.4%estimated ± 5.1 pp, low confidence
10GPT-6.1 Sol36.2%measured
11Claude Opus 4.836.1%estimated ± 5.1 pp, low confidence
12GPT-5.536.1%estimated ± 5.1 pp, low confidence
13Kimi K336.0%estimated ± 5.1 pp, low confidence
14Claude Opus 535.5%estimated ± 0.6 pp, medium confidence
15Inkling 2 Small35.1%estimated ± 5.1 pp, low confidence
16GPT-6.2 Astra34.6%estimated ± 1.2 pp, medium confidence
17Ornith-1.5-397B34.4%estimated ± 5.1 pp, low confidence
18Claude Fable 5.234.4%estimated ± 8.6 pp, low confidence
19Muse Wonder Full34.2%estimated ± 8.6 pp, low confidence
20GPT-6.2 Sol34.2%estimated ± 1.2 pp, medium confidence
21Claude Sonnet 5.534.1%estimated ± 8.6 pp, low confidence
22GPT-5.5 Pro34.0%estimated ± 8.6 pp, low confidence
23GPT-5.4 Pro33.9%estimated ± 8.6 pp, low confidence
24Claude Fable 5.133.9%estimated ± 8.6 pp, low confidence
25Claude Fable 533.8%estimated ± 8.6 pp, low confidence
26Muse Wonder33.8%estimated ± 8.6 pp, low confidence
27Gemini 4 Argon33.6%estimated ± 8.6 pp, low confidence
28MiMo-V2.6-Pro33.6%estimated ± 8.6 pp, low confidence
29Qwen3.8 Max33.4%estimated ± 5.1 pp, low confidence
30Muse Spark 1.333.4%estimated ± 8.6 pp, low confidence
31GPT-5.6 Sol33.1%measured
32MiniMax M3.1 Flash33.0%estimated ± 8.6 pp, low confidence
33Step 5 Preview32.7%estimated ± 8.6 pp, low confidence
34GPT-5.6 Terra32.7%measured
35Sage-A1.532.5%estimated ± 8.6 pp, low confidence
36GLM-5.332.3%estimated ± 8.6 pp, low confidence
37Claude Haiku 5.532.3%estimated ± 8.6 pp, low confidence
38Gemini 3.8 Flash32.2%estimated ± 8.6 pp, low confidence
39Ling 3.1 Flash32.1%estimated ± 8.6 pp, low confidence
40GPT-5.232.1%estimated ± 5.1 pp, low confidence
41Qwen3.7 Max32.1%estimated ± 5.1 pp, low confidence
42Grok 4.732.0%estimated ± 8.6 pp, low confidence
43Muse Spark 1.232.0%estimated ± 8.6 pp, low confidence
44Qwen3.8 Max Preview32.0%estimated ± 8.6 pp, low confidence
45GPT-5.6 Luna32.0%measured
46Beam 1.0 501B-A23B FP831.9%estimated ± 8.6 pp, low confidence
47Grok 4.631.9%estimated ± 8.6 pp, low confidence
48Claude Sonnet 531.8%estimated ± 8.6 pp, low confidence
49GPT-5.3 Codex31.8%estimated ± 8.6 pp, low confidence
50Beam 1.0 501B-A23B31.6%estimated ± 8.6 pp, low confidence
51GPT-6.1 Terra31.4%measured
52GPT-6 Luna31.4%measured
53Grok 4.531.4%estimated ± 8.6 pp, low confidence
54Beam 1.0 501B-A23B NVFP431.4%estimated ± 8.6 pp, low confidence
55Muse Spark 1.131.3%estimated ± 8.6 pp, low confidence
56Hy4 preview31.2%estimated ± 5.1 pp, low confidence
57Muse Spark Open31.2%estimated ± 8.6 pp, low confidence
58Gemini 3.7 Flash31.0%estimated ± 8.6 pp, low confidence
59Claude Opus 4.6 (Adaptive)30.3%estimated ± 8.6 pp, low confidence
60Gemini 3.5 Flash30.2%estimated ± 5.1 pp, low confidence
61GPT-6 Sol30.1%measured
62MiMo-V2.6-Flash30.0%estimated ± 8.6 pp, low confidence
63Nemotron 3.5 Super VL 120B A12B29.4%estimated ± 8.6 pp, low confidence
64Gemini 3.6 Flash29.3%estimated ± 8.6 pp, low confidence
65Nemotron 3.5 Super VL 120B A12B NVFP429.3%estimated ± 8.6 pp, low confidence
66Kimi K2.7 Code28.9%estimated ± 8.6 pp, low confidence
67Mistral Large 428.9%estimated ± 8.6 pp, low confidence
68Quasar 438B28.5%estimated ± 8.6 pp, low confidence
69GPT-5.4 nano28.5%estimated ± 8.6 pp, low confidence
70Gemini 3 Pro28.3%estimated ± 8.6 pp, low confidence
71A.X K228.2%estimated ± 8.6 pp, low confidence
72GPT-5.2-Codex28.0%estimated ± 8.6 pp, low confidence
73Solar Pro 425.6%estimated ± 8.6 pp, low confidence
74GPT-524.8%estimated ± 8.6 pp, low confidence
75GPT-5.1-Codex24.8%estimated ± 8.6 pp, low confidence
76GPT-5.1-Codex-Max24.8%estimated ± 8.6 pp, low confidence
77GPT-5.123.5%estimated ± 8.6 pp, low confidence
78Hy323.5%estimated ± 8.6 pp, low confidence
79Qwen3.8-Flash-Next23.3%estimated ± 5.1 pp, low confidence
80Apodex 1.123.0%estimated ± 8.6 pp, low confidence
81Apodex 1.1 Mini23.0%estimated ± 8.6 pp, low confidence
82Claude Opus 4.5 Thinking23.0%estimated ± 8.6 pp, low confidence
83GLM-5.123.0%estimated ± 8.6 pp, low confidence
84Nemotron 3.5 Bolt 74B-A9B22.4%estimated ± 8.6 pp, low confidence
85MiMo-V2.5-Pro21.7%estimated ± 8.6 pp, low confidence
86MiniMax M321.0%estimated ± 8.6 pp, low confidence
87Qwen 3.6 Max (preview)21.0%estimated ± 8.6 pp, low confidence
88Gemini 3.1 Pro20.6%measured
89Grok 4.2020.3%measured
90Grok 4.1 Fast (Reasoning)18.8%estimated ± 8.6 pp, low confidence
91Grok 4 Fast (Reasoning)18.8%estimated ± 8.6 pp, low confidence
92Gemini 2.5 Pro17.7%estimated ± 8.6 pp, low confidence
93Muse Glimmer 30B17.7%estimated ± 8.6 pp, low confidence
94Step 3.7 Flash16.6%estimated ± 8.6 pp, low confidence
95GLM-515.3%estimated ± 8.6 pp, low confidence
96Grok 415.3%estimated ± 8.6 pp, low confidence
97Ling 3.0 Flash VL15.3%estimated ± 8.6 pp, low confidence
98Claude Opus 4.614.8%measured
99GLM-5.214.3%estimated ± 5.1 pp, low confidence
100GLM-4.713.9%estimated ± 8.6 pp, low confidence
101Ling 3.0 Flash13.9%estimated ± 8.6 pp, low confidence
102DeepSeek-R112.3%estimated ± 8.6 pp, low confidence
103Gemini 2.5 Flash12.3%estimated ± 8.6 pp, low confidence
104Gemini 3 Flash12.3%estimated ± 8.6 pp, low confidence
105Gemma 4 31B12.3%estimated ± 8.6 pp, low confidence
106GPT-OSS 20B12.3%estimated ± 8.6 pp, low confidence
107Qwen3.8-Omni-Flash11.0%estimated ± 5.1 pp, low confidence
108Claude 4 Sonnet10.4%estimated ± 8.6 pp, low confidence
109GPT-OSS 120B10.4%estimated ± 8.6 pp, low confidence
110K-Exaone10.4%estimated ± 8.6 pp, low confidence
111MiMo-V2-Omni10.4%estimated ± 8.6 pp, low confidence
112o310.4%estimated ± 8.6 pp, low confidence
113DeepSeek V4.1 Flash 10179.5%estimated ± 5.1 pp, low confidence
114Claude 3.7 Sonnet Thinking8.9%estimated ± 8.6 pp, low confidence
115DeepSeek V3.28.9%estimated ± 8.6 pp, low confidence
116K-EXAONE 2.08.9%estimated ± 8.6 pp, low confidence
117Qwen3.5-27B8.9%estimated ± 8.6 pp, low confidence
118Qwen3.5-35B-A3B8.9%estimated ± 8.6 pp, low confidence
119Trinity-Large-Preview8.9%estimated ± 8.6 pp, low confidence
120Trinity-Large-Thinking8.9%estimated ± 8.6 pp, low confidence
121Gemma 4 E4B6.5%estimated ± 8.6 pp, low confidence
122GLM-5V-Turbo6.5%estimated ± 8.6 pp, low confidence
123MiniMax M2.76.5%estimated ± 8.6 pp, low confidence
124o4-mini6.5%estimated ± 8.6 pp, low confidence
125Qwen3.5-122B-A10B6.5%estimated ± 8.6 pp, low confidence
126Kimi K2.65.0%estimated ± 5.1 pp, low confidence
127Qwen3.6 Plus4.2%estimated ± 5.1 pp, low confidence
128Command A+3.5%estimated ± 8.6 pp, low confidence
129GLM-5-Turbo3.5%estimated ± 8.6 pp, low confidence
130Granite 4.2 30B3.5%estimated ± 8.6 pp, low confidence
131Granite 4.2 8B3.5%estimated ± 8.6 pp, low confidence
132Magistral Medium3.5%estimated ± 8.6 pp, low confidence
133MiMo-V2-Pro3.5%estimated ± 8.6 pp, low confidence
134MiniCPM5-2B3.5%estimated ± 8.6 pp, low confidence
135Mistral Small 43.5%estimated ± 8.6 pp, low confidence
136North Mini Code3.5%estimated ± 8.6 pp, low confidence
137o13.5%estimated ± 8.6 pp, low confidence
138Qwen3.6-35B-A3B3.5%estimated ± 8.6 pp, low confidence
139Sarvam 30B3.5%estimated ± 8.6 pp, low confidence
140Qwen3.7 Plus3.5%estimated ± 5.1 pp, low confidence
141DeepSeek V4 Pro2.4%estimated ± 5.1 pp, low confidence
142Grok 4.32.4%estimated ± 5.1 pp, low confidence
143Claude Sonnet 4.61.7%estimated ± 5.1 pp, low confidence
144Interfaze Beta1.7%estimated ± 5.1 pp, low confidence
145Inkling-Small0.8%estimated ± 5.1 pp, low confidence
146Ornith-1.5-35B-A3B0.4%estimated ± 5.1 pp, low confidence
147Qwen3.8-27B0.4%estimated ± 5.1 pp, low confidence
148Qwen3.5 397B0.1%estimated ± 5.1 pp, low confidence
149DeepSeek V4 Flash 07310.0%estimated ± 5.1 pp, low confidence
150GPT-5.4 mini0.0%estimated ± 5.1 pp, low confidence
151Inkling0.0%estimated ± 5.1 pp, low confidence
152Qwen3.6-27B0.0%estimated ± 5.1 pp, low confidence
153Kimi K2.50.0%estimated ± 5.1 pp, low confidence
154Hy3 Preview0.0%estimated ± 5.1 pp, low confidence
155Claude Opus 4.50.0%estimated ± 5.1 pp, low confidence
156Nemotron 3 Ultra0.0%estimated ± 5.1 pp, low confidence
157Celeris-10.0%estimated ± 8.6 pp, low confidence
158Claude 3.5 Haiku0.0%estimated ± 8.6 pp, low confidence
159Claude 3.7 Sonnet0.0%estimated ± 8.6 pp, low confidence
160Claude 3 Haiku0.0%estimated ± 8.6 pp, low confidence
161DeepSeek V30.0%estimated ± 8.6 pp, low confidence
162DeepSeek V3 03240.0%estimated ± 8.6 pp, low confidence
163DeepSeek V3.10.0%estimated ± 8.6 pp, low confidence
164Devstral Small0.0%estimated ± 8.6 pp, low confidence
165Exaone 4.0 1.2B0.0%estimated ± 8.6 pp, low confidence
166Exaone 4.0 32B0.0%estimated ± 8.6 pp, low confidence
167Gemini 2.0 Flash0.0%estimated ± 8.6 pp, low confidence
168Gemini 2.5 Flash-Lite0.0%estimated ± 8.6 pp, low confidence
169Gemini 3.5 Flash-Lite0.0%estimated ± 8.6 pp, low confidence
170Gemma 3 27B0.0%estimated ± 8.6 pp, low confidence
171Gemma 4 12B0.0%estimated ± 8.6 pp, low confidence
172Gemma 4 26B A4B0.0%estimated ± 8.6 pp, low confidence
173Gemma 4 E2B0.0%estimated ± 8.6 pp, low confidence
174GLM-4.5-Air0.0%estimated ± 8.6 pp, low confidence
175GLM-4.60.0%estimated ± 8.6 pp, low confidence
176GPT-4.10.0%estimated ± 8.6 pp, low confidence
177GPT-4.1 mini0.0%estimated ± 8.6 pp, low confidence
178GPT-4.1 nano0.0%estimated ± 8.6 pp, low confidence
179GPT-4o0.0%estimated ± 8.6 pp, low confidence
180Granite-4.0-350M0.0%estimated ± 8.6 pp, low confidence
181Granite-4.0-H-1B0.0%estimated ± 8.6 pp, low confidence
182Granite-4.0-H-350M0.0%estimated ± 8.6 pp, low confidence
183Granite 4.2 3B0.0%estimated ± 8.6 pp, low confidence
184Grok 4.1 Fast0.0%estimated ± 8.6 pp, low confidence
185Grok Code Fast 10.0%estimated ± 8.6 pp, low confidence
186Kimi K20.0%estimated ± 8.6 pp, low confidence
187LFM2.5-2.6B0.0%estimated ± 8.6 pp, low confidence
188LFM2.5-8B-A1B0.0%estimated ± 8.6 pp, low confidence
189LFM2.5-VL-1.6B-Extract0.0%estimated ± 8.6 pp, low confidence
190Ling 2.6 Flash0.0%estimated ± 8.6 pp, low confidence
191Ling 3.0 Tiny0.0%estimated ± 8.6 pp, low confidence
192Llama 3.1 405B0.0%estimated ± 8.6 pp, low confidence
193Llama 3.1 70B0.0%estimated ± 8.6 pp, low confidence
194Llama 3.3 70B0.0%estimated ± 8.6 pp, low confidence
195Llama 4 Maverick0.0%estimated ± 8.6 pp, low confidence
196Llama 4 Scout0.0%estimated ± 8.6 pp, low confidence
197Magistral Small0.0%estimated ± 8.6 pp, low confidence
198Mercury 2.50.0%estimated ± 8.6 pp, low confidence
199MiMo-V2-Flash0.0%estimated ± 8.6 pp, low confidence
200Mistral Large 20.0%estimated ± 8.6 pp, low confidence
201Mistral Large 30.0%estimated ± 8.6 pp, low confidence
202Mistral Medium 30.0%estimated ± 8.6 pp, low confidence
203Mistral Medium 3.5 128B0.0%estimated ± 8.6 pp, low confidence
204Mistral Small 3.10.0%estimated ± 8.6 pp, low confidence
205Nemotron 3.5 Lightning 30B A3B NVFP40.0%estimated ± 8.6 pp, low confidence
206Nemotron 3 Nano 30B0.0%estimated ± 8.6 pp, low confidence
207Nemotron 3 Nano Omni 30B A3B0.0%estimated ± 8.6 pp, low confidence
208Nemotron Ultra 253B0.0%estimated ± 8.6 pp, low confidence
209Nova Pro0.0%estimated ± 8.6 pp, low confidence
210Phi-40.0%estimated ± 8.6 pp, low confidence
211Phi-4-reasoning0.0%estimated ± 8.6 pp, low confidence
212Qwen3.5-4B0.0%estimated ± 8.6 pp, low confidence
213Qwen3 Max0.0%estimated ± 8.6 pp, low confidence
214Qwen3-Omni-30B-A3B-Instruct0.0%estimated ± 8.6 pp, low confidence
215Qwen3-Omni-30B-A3B-Thinking0.0%estimated ± 8.6 pp, low confidence
216Sarvam 105B0.0%estimated ± 8.6 pp, low confidence
217Solar Pro 20.0%estimated ± 8.6 pp, low confidence
218Solar Pro 30.0%estimated ± 8.6 pp, low confidence
219Ultravox v0.6 Llama 3.3 70B0.0%estimated ± 8.6 pp, low confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General