Knowledge & reasoning

BioMysteryBench (human-difficult) leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-07, the highest measured score on BioMysteryBench (human-difficult) is 56.5% by Gemini 3.8 Flash. 197 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Gemini 3.8 Flash	56.5%	measured
2	Claude Fable 5.1	51.6%	estimated ± 5.9 pp, low confidence
3	Claude Fable 5	51.1%	estimated ± 5.9 pp, low confidence
4	GPT-6 Astra	50.2%	estimated ± 5.9 pp, low confidence
5	GPT-6.1 Sol	50.1%	estimated ± 5.9 pp, low confidence
6	Claude Opus 5.5	50.0%	measured
7	GPT-5.5 Pro	49.8%	estimated ± 6.8 pp, low confidence
8	GPT-5.4 Pro	49.7%	estimated ± 6.8 pp, low confidence
9	Claude Opus 5	49.4%	measured
10	GPT-5.6 Sol	49.2%	estimated ± 5.9 pp, low confidence
11	Gemini 3 Pro Deep Think	49.1%	estimated ± 6.8 pp, low confidence
12	GPT-5.5	48.7%	estimated ± 5.9 pp, low confidence
13	Gemini 3 Pro	47.9%	estimated ± 5.9 pp, low confidence
14	Claude Haiku 5.5	47.8%	estimated ± 6.8 pp, low confidence
15	Gemini 3.1 Pro	47.5%	estimated ± 5.9 pp, low confidence
16	GPT-6 Sol	47.4%	estimated ± 5.9 pp, low confidence
17	Hy4 preview	47.2%	estimated ± 6.8 pp, low confidence
18	GPT-5.3 Codex	46.8%	estimated ± 5.9 pp, low confidence
19	GLM-5.3-Flash	46.7%	estimated ± 6.8 pp, low confidence
20	Muse Spark 1.1	46.5%	estimated ± 5.9 pp, low confidence
21	Gemini 3.5 Flash	46.4%	estimated ± 5.9 pp, low confidence
22	Grok 4.5	46.3%	estimated ± 5.9 pp, low confidence
23	GPT-5.4	45.9%	estimated ± 5.9 pp, low confidence
24	Gemini 3.6 Flash	45.6%	estimated ± 5.9 pp, low confidence
25	Gemini 4 Argon	45.6%	estimated ± 5.9 pp, low confidence
26	Muse Spark	45.4%	estimated ± 5.9 pp, low confidence
27	DeepSeek V4 Pro 0813	45.2%	estimated ± 5.9 pp, low confidence
28	Claude Opus 4.7 (Adaptive)	45.1%	estimated ± 5.9 pp, low confidence
29	Claude Opus 4.8	45.1%	estimated ± 5.9 pp, low confidence
30	Grok 4.6	44.8%	estimated ± 5.9 pp, low confidence
31	Claude Sonnet 5.5	44.7%	measured
32	Kimi K3	44.6%	estimated ± 5.9 pp, low confidence
33	Grok 4.7	44.5%	estimated ± 5.9 pp, low confidence
34	Claude Opus 4.6 (Adaptive)	44.3%	estimated ± 5.9 pp, low confidence
35	GPT-5.6 Terra	44.2%	estimated ± 5.9 pp, low confidence
36	Claude Opus 4.5 Thinking	44.1%	estimated ± 5.9 pp, low confidence
37	DeepSeek V4.1 Flash	44.1%	estimated ± 5.9 pp, low confidence
38	Claude Opus 4.6	43.8%	estimated ± 5.9 pp, low confidence
39	Gemini 3 Flash	43.8%	estimated ± 5.9 pp, low confidence
40	Muse Spark 1.2	43.6%	estimated ± 5.9 pp, low confidence
41	Gemini 3.7 Flash	43.5%	measured
42	Claude Opus 4.7	43.3%	estimated ± 5.9 pp, low confidence
43	GPT-5.2	43.1%	estimated ± 5.9 pp, low confidence
44	GPT-6 Luna	42.8%	estimated ± 5.9 pp, low confidence
45	Muse Spark 1.3	42.7%	estimated ± 5.9 pp, low confidence
46	GPT-5.6 Luna	42.3%	estimated ± 5.9 pp, low confidence
47	Inkling	41.8%	estimated ± 5.9 pp, low confidence
48	Step 5 Preview	41.7%	estimated ± 5.9 pp, low confidence
49	GPT-5.2-Codex	41.5%	estimated ± 5.9 pp, low confidence
50	Claude Opus 4.5	41.4%	estimated ± 5.9 pp, low confidence
51	Grok 4	41.2%	estimated ± 5.9 pp, low confidence
52	DeepSeek V4 Flash 0731	41.1%	estimated ± 5.9 pp, low confidence
53	GPT-5 (high)	41.1%	estimated ± 5.9 pp, low confidence
54	Claude Sonnet 5	41.0%	estimated ± 5.9 pp, low confidence
55	GPT-5.1-Codex	40.9%	estimated ± 5.9 pp, low confidence
56	GPT-5.1-Codex-Max	40.9%	estimated ± 5.9 pp, low confidence
57	Kimi K2.7 Code	40.7%	estimated ± 5.9 pp, low confidence
58	GPT-5 (medium)	40.7%	estimated ± 5.9 pp, low confidence
59	Gemini 2.5 Pro	40.5%	estimated ± 5.9 pp, low confidence
60	Claude Sonnet 4.6	40.2%	estimated ± 5.9 pp, low confidence
61	o3	40.2%	estimated ± 5.9 pp, low confidence
62	Qwen 3.6 Max (preview)	39.8%	estimated ± 5.9 pp, low confidence
63	GPT-5.1	39.7%	estimated ± 5.9 pp, low confidence
64	GPT-5.4 mini	39.6%	estimated ± 5.9 pp, low confidence
65	Kimi K2.5	38.3%	estimated ± 5.9 pp, low confidence
66	Kimi K2.5 (Reasoning)	38.3%	estimated ± 5.9 pp, low confidence
67	MiMo-V2.6-Pro	38.0%	estimated ± 5.9 pp, low confidence
68	Grok 4.3	37.9%	estimated ± 5.9 pp, low confidence
69	o1	37.8%	estimated ± 5.9 pp, low confidence
70	GLM-5.3	37.5%	estimated ± 5.9 pp, low confidence
71	Inkling-Small	37.1%	estimated ± 5.9 pp, low confidence
72	Kimi K2.6	36.7%	estimated ± 5.9 pp, low confidence
73	Hy3	36.3%	estimated ± 5.9 pp, low confidence
74	Apodex 1.1 Mini	36.1%	estimated ± 5.9 pp, low confidence
75	Qwen3.8 Max Preview	36.1%	estimated ± 5.9 pp, low confidence
76	Hy3 Preview	36.0%	estimated ± 5.9 pp, low confidence
77	Qwen3.7 Max	35.7%	estimated ± 5.9 pp, low confidence
78	DeepSeek-R1	35.3%	estimated ± 5.9 pp, low confidence
79	Apodex 1.1	35.3%	measured
80	Claude 3 Opus	35.2%	estimated ± 8.1 pp, low confidence
81	Claude 4.1 Opus	35.2%	estimated ± 8.1 pp, low confidence
82	DeepSeek R1 Distill Qwen 32B	35.2%	estimated ± 8.1 pp, low confidence
83	Gemini 1.0 Pro	35.2%	estimated ± 8.1 pp, low confidence
84	Gemini 1.5 Pro	35.2%	estimated ± 8.1 pp, low confidence
85	GPT-4 Turbo	35.2%	estimated ± 8.1 pp, low confidence
86	GPT-4o mini	35.2%	estimated ± 8.1 pp, low confidence
87	o1-preview	35.2%	estimated ± 8.1 pp, low confidence
88	o1-pro	35.2%	estimated ± 8.1 pp, low confidence
89	o3-mini	35.2%	estimated ± 8.1 pp, low confidence
90	o3-pro	35.2%	estimated ± 8.1 pp, low confidence
91	Phi-4 Multimodal Instruct	35.2%	estimated ± 8.1 pp, low confidence
92	Qwen2.5 Coder 32B Instruct	35.2%	estimated ± 8.1 pp, low confidence
93	Gemini 3.5 Flash-Lite	34.6%	estimated ± 5.9 pp, low confidence
94	GLM-4.7	34.5%	estimated ± 5.9 pp, low confidence
95	GLM-5V-Turbo	34.5%	estimated ± 5.9 pp, low confidence
96	Ling 3.1 Flash	34.3%	estimated ± 5.9 pp, low confidence
97	DeepSeek V3.1 (Reasoning)	34.3%	estimated ± 5.9 pp, low confidence
98	GLM-5-Turbo	33.9%	estimated ± 5.9 pp, low confidence
99	GPT-4.1	33.4%	estimated ± 5.9 pp, low confidence
100	Kimi K2	33.1%	estimated ± 5.9 pp, low confidence
101	MiMo-V2.6-Flash	32.8%	estimated ± 5.9 pp, low confidence
102	Muse Glimmer 30B	32.8%	estimated ± 5.9 pp, low confidence
103	MiniMax M2.7	32.7%	estimated ± 5.9 pp, low confidence
104	MiMo-V2-Pro	32.5%	estimated ± 5.9 pp, low confidence
105	Qwen3.6 Plus	32.4%	estimated ± 5.9 pp, low confidence
106	GLM-5	32.3%	estimated ± 5.9 pp, low confidence
107	Gemini 2.5 Flash	32.2%	estimated ± 5.9 pp, low confidence
108	Mistral Large 4	31.9%	estimated ± 5.9 pp, low confidence
109	Step 3.7 Flash	31.9%	estimated ± 5.9 pp, low confidence
110	GPT-5.4 nano	31.9%	estimated ± 5.9 pp, low confidence
111	DeepSeek V3	31.7%	estimated ± 5.9 pp, low confidence
112	Grok 4.1 Fast (Reasoning)	31.4%	estimated ± 5.9 pp, low confidence
113	Mistral Large 3	31.3%	estimated ± 5.9 pp, low confidence
114	Llama 4 Maverick	31.2%	estimated ± 5.9 pp, low confidence
115	Mistral Medium 3.5 128B	31.1%	estimated ± 5.9 pp, low confidence
116	Qwen3.5 397B	30.9%	estimated ± 5.9 pp, low confidence
117	Qwen3.5 397B (Reasoning)	30.9%	estimated ± 5.9 pp, low confidence
118	Qwen3.8-Flash-Next	30.9%	estimated ± 5.9 pp, low confidence
119	Qwen3.5-122B-A10B	30.8%	estimated ± 5.9 pp, low confidence
120	Qwen3 Max	30.8%	estimated ± 5.9 pp, low confidence
121	DeepSeek V3 0324	30.8%	estimated ± 5.9 pp, low confidence
122	GLM-5.2	30.8%	estimated ± 5.9 pp, low confidence
123	Nemotron 3 Super 100B	30.8%	estimated ± 5.9 pp, low confidence
124	DeepSeek V3.2	30.5%	estimated ± 5.9 pp, low confidence
125	GLM-5.1	30.3%	estimated ± 5.9 pp, low confidence
126	Grok Code Fast 1	30.1%	estimated ± 5.9 pp, low confidence
127	Llama 3.1 405B	29.9%	estimated ± 5.9 pp, low confidence
128	DeepSeek V3.1	29.8%	estimated ± 5.9 pp, low confidence
129	Grok 4 Fast (Reasoning)	29.5%	estimated ± 5.9 pp, low confidence
130	Claude 4 Sonnet	29.4%	estimated ± 5.9 pp, low confidence
131	Qwen3.7 Plus	29.3%	estimated ± 5.9 pp, low confidence
132	Trinity-Large-Preview	29.3%	estimated ± 5.9 pp, low confidence
133	Trinity-Large-Thinking	29.3%	estimated ± 5.9 pp, low confidence
134	MiMo-V2.5-Pro	29.2%	estimated ± 5.9 pp, low confidence
135	Mercury 2.5	28.8%	estimated ± 5.9 pp, low confidence
136	GPT-OSS 120B	28.7%	estimated ± 5.9 pp, low confidence
137	Mistral Small 4	28.6%	estimated ± 5.9 pp, low confidence
138	Mistral Small 4 (Reasoning)	28.6%	estimated ± 5.9 pp, low confidence
139	Nemotron 3 Ultra	28.5%	estimated ± 5.9 pp, low confidence
140	GLM-4.6	28.3%	estimated ± 5.9 pp, low confidence
141	Qwen3.5-27B	27.7%	estimated ± 5.9 pp, low confidence
142	GPT-4.1 mini	27.3%	estimated ± 5.9 pp, low confidence
143	Nemotron Ultra 253B	27.2%	estimated ± 5.9 pp, low confidence
144	Qwen3.5-35B-A3B	27.2%	estimated ± 5.9 pp, low confidence
145	Gemma 4 31B	27.1%	estimated ± 5.9 pp, low confidence
146	GPT-4o	27.0%	estimated ± 5.9 pp, low confidence
147	Mistral Large 2	27.0%	estimated ± 5.9 pp, low confidence
148	Qwen3.6-27B	26.7%	estimated ± 5.9 pp, low confidence
149	MiMo-V2-Omni	26.4%	estimated ± 5.9 pp, low confidence
150	Gemma 4 26B A4B	26.2%	estimated ± 5.9 pp, low confidence
151	Ultravox v0.6 Llama 3.3 70B	26.1%	estimated ± 5.9 pp, low confidence
152	North Mini Code	26.0%	estimated ± 5.9 pp, low confidence
153	Solar Pro 4	26.0%	estimated ± 5.9 pp, low confidence
154	Qwen3.6-35B-A3B	25.9%	estimated ± 5.9 pp, low confidence
155	A.X K2	25.8%	estimated ± 5.9 pp, low confidence
156	Solar Pro 3	25.7%	estimated ± 5.9 pp, low confidence
157	Mistral Medium 3	25.5%	estimated ± 5.9 pp, low confidence
158	Ling 3.0 Flash	25.4%	estimated ± 5.9 pp, low confidence
159	Ling 3.0 Flash FP8	25.4%	estimated ± 5.9 pp, low confidence
160	Claude 3 Haiku	24.8%	estimated ± 5.9 pp, low confidence
161	Sarvam 105B	24.8%	estimated ± 5.9 pp, low confidence
162	Nemotron 3 Nano 30B	24.5%	estimated ± 5.9 pp, low confidence
163	Grok 4.1 Fast	24.4%	estimated ± 5.9 pp, low confidence
164	Nova Pro	24.1%	estimated ± 5.9 pp, low confidence
165	MiniMax M3	23.9%	estimated ± 5.9 pp, low confidence
166	K-Exaone	23.6%	estimated ± 5.9 pp, low confidence
167	GLM-4.5-Air	23.5%	estimated ± 5.9 pp, low confidence
168	Solar Pro 2	23.3%	estimated ± 5.9 pp, low confidence
169	GPT-OSS 20B	23.2%	estimated ± 5.9 pp, low confidence
170	Gemma 4 12B	22.7%	estimated ± 5.9 pp, low confidence
171	Ling 2.6 Flash	22.7%	estimated ± 5.9 pp, low confidence
172	MiMo-V2-Flash	22.7%	estimated ± 5.9 pp, low confidence
173	Qwen3.8-27B	22.7%	estimated ± 5.9 pp, low confidence
174	Quasar 438B	22.6%	estimated ± 5.9 pp, low confidence
175	Llama 4 Scout	22.3%	estimated ± 5.9 pp, low confidence
176	Nemotron 3 Nano Omni 30B A3B	22.3%	estimated ± 5.9 pp, low confidence
177	Qwen3-Omni-30B-A3B-Thinking	21.6%	estimated ± 5.9 pp, low confidence
178	Ling 3.0 Flash VL	21.4%	estimated ± 5.9 pp, low confidence
179	Nemotron 3.5 Lightning 30B A3B NVFP4	21.4%	estimated ± 5.9 pp, low confidence
180	Qwen3-Omni-30B-A3B-Instruct	21.3%	estimated ± 5.9 pp, low confidence
181	Phi-4	21.1%	estimated ± 5.9 pp, low confidence
182	GPT-4.1 nano	20.6%	estimated ± 5.9 pp, low confidence
183	K-EXAONE 2.0	20.0%	estimated ± 5.9 pp, low confidence
184	Gemma 3 27B	19.8%	estimated ± 5.9 pp, low confidence
185	Sarvam 30B	19.4%	estimated ± 5.9 pp, low confidence
186	Granite 4.2 8B	17.7%	estimated ± 5.9 pp, low confidence
187	Celeris-1	17.4%	estimated ± 5.9 pp, low confidence
188	Exaone 4.0 32B	16.9%	estimated ± 5.9 pp, low confidence
189	Granite 4.2 30B	16.3%	estimated ± 5.9 pp, low confidence
190	LFM2.5-8B-A1B	15.3%	estimated ± 5.9 pp, low confidence
191	Granite 4.2 3B	15.1%	estimated ± 5.9 pp, low confidence
192	Command A+	14.7%	estimated ± 5.9 pp, low confidence
193	Gemma 4 E4B	14.3%	estimated ± 5.9 pp, low confidence
194	Ling 3.0 Tiny	14.1%	estimated ± 5.9 pp, low confidence
195	MiniCPM5-2B	14.0%	estimated ± 5.9 pp, low confidence
196	Gemma 4 E2B	11.4%	estimated ± 5.9 pp, low confidence
197	LFM2.5-VL-1.6B-Extract	10.2%	estimated ± 5.9 pp, low confidence
198	Granite-4.0-H-1B	9.2%	estimated ± 5.9 pp, low confidence
199	Exaone 4.0 1.2B	8.9%	estimated ± 5.9 pp, low confidence
200	LFM2.5-2.6B	8.0%	estimated ± 5.9 pp, low confidence
201	Granite-4.0-350M	7.1%	estimated ± 5.9 pp, low confidence
202	Granite-4.0-H-350M	7.0%	estimated ± 5.9 pp, low confidence
203	Claude 4.1 Opus Thinking	0.0%	estimated ± 6.8 pp, low confidence

BioMysteryBench (human-difficult) leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General