Knowledge & reasoning

STEMBench leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on STEMBench is 2.8% by GPT-6 Astra. 208 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6 Astra	2.8%	measured
2	GPT-6.1 Sol	2.8%	measured
3	Gemini 4 Carbon	2.7%	estimated ± 0.2 pp, medium confidence
4	Muse Wonder Full	2.7%	estimated ± 0.2 pp, medium confidence
5	GPT-5.6 Sol	2.7%	estimated ± 0.2 pp, medium confidence
6	Claude Sonnet 5.5	2.7%	estimated ± 0.2 pp, high confidence
7	GPT-6 Sol	2.7%	estimated ± 0.2 pp, high confidence
8	GPT-5.5 Pro	2.7%	estimated ± 0.2 pp, high confidence
9	GPT-5.4 Pro	2.7%	estimated ± 0.2 pp, high confidence
10	GPT-5.6 Terra	2.7%	estimated ± 0.2 pp, high confidence
11	Claude Fable 5.1	2.7%	estimated ± 0.2 pp, high confidence
12	Claude Opus 5	2.7%	estimated ± 0.2 pp, high confidence
13	Muse Wonder	2.7%	measured
14	Claude Fable 5	2.7%	estimated ± 0.2 pp, high confidence
15	Claude Opus 5.5	2.6%	measured
16	Hy4 preview	2.6%	measured
17	GPT-6.1 Terra	2.6%	estimated ± 0.2 pp, high confidence
18	DeepSeek V4.1 Flash 1017	2.5%	measured
19	Mistral Large 4	2.5%	measured
20	GPT-5.5	2.5%	estimated ± 0.2 pp, high confidence
21	Gemini 3 Pro Deep Think	2.5%	estimated ± 0.2 pp, high confidence
22	Muse Spark 1.3	2.4%	estimated ± 0.2 pp, high confidence
23	Kimi K3.1 Pro	2.4%	estimated ± 0.2 pp, high confidence
24	GPT-5.4	2.4%	estimated ± 0.2 pp, high confidence
25	Kimi K3	2.4%	estimated ± 0.2 pp, high confidence
26	MiniMax M3.1 Flash	2.4%	estimated ± 0.2 pp, high confidence
27	Claude Opus 4.8	2.4%	estimated ± 0.2 pp, high confidence
28	GLM-5.2	2.4%	estimated ± 0.2 pp, high confidence
29	Step 5 Preview	2.4%	estimated ± 0.2 pp, high confidence
30	GPT-5.6 Luna	2.4%	estimated ± 0.2 pp, high confidence
31	Sage-A1.5	2.4%	estimated ± 0.2 pp, high confidence
32	GPT-6 Luna	2.4%	estimated ± 0.2 pp, high confidence
33	GLM-5.3	2.4%	estimated ± 0.2 pp, high confidence
34	Claude Haiku 5.5	2.4%	estimated ± 0.2 pp, high confidence
35	Gemini 3.8 Flash	2.4%	estimated ± 0.2 pp, high confidence
36	DeepSeek V4 Pro 0813	2.4%	estimated ± 0.2 pp, high confidence
37	Ling 3.1 Flash	2.4%	estimated ± 0.2 pp, high confidence
38	Gemini 3.1 Pro	2.4%	estimated ± 0.2 pp, high confidence
39	Grok 4.7	2.4%	estimated ± 0.2 pp, high confidence
40	Muse Spark 1.2	2.4%	estimated ± 0.2 pp, high confidence
41	Qwen3.8 Max Preview	2.4%	estimated ± 0.2 pp, high confidence
42	Beam 1.0 501B-A23B FP8	2.4%	estimated ± 0.2 pp, high confidence
43	Grok 4.6	2.4%	estimated ± 0.2 pp, high confidence
44	Claude Sonnet 5	2.4%	estimated ± 0.2 pp, high confidence
45	GPT-5.3 Codex	2.4%	estimated ± 0.2 pp, high confidence
46	DeepSeek V4 Flash 0731	2.4%	estimated ± 0.2 pp, high confidence
47	Beam 1.0 501B-A23B	2.4%	estimated ± 0.2 pp, high confidence
48	Grok 4.5	2.4%	estimated ± 0.2 pp, high confidence
49	Beam 1.0 501B-A23B NVFP4	2.4%	estimated ± 0.2 pp, high confidence
50	Muse Spark 1.1	2.4%	estimated ± 0.2 pp, high confidence
51	Muse Spark Open	2.4%	estimated ± 0.2 pp, high confidence
52	Gemini 3.7 Flash	2.4%	estimated ± 0.2 pp, high confidence
53	Qwen3.7 Max	2.4%	estimated ± 0.2 pp, high confidence
54	Gemini 3.5 Flash	2.4%	estimated ± 0.2 pp, high confidence
55	Claude Opus 4.6 (Adaptive)	2.4%	estimated ± 0.2 pp, high confidence
56	Claude Opus 4.7 (Adaptive)	2.4%	estimated ± 0.2 pp, high confidence
57	MiMo-V2.6-Flash	2.4%	estimated ± 0.2 pp, high confidence
58	GPT-5.2	2.4%	estimated ± 0.2 pp, high confidence
59	Muse Spark	2.4%	estimated ± 0.2 pp, high confidence
60	Nemotron 3.5 Super VL 120B A12B	2.4%	estimated ± 0.2 pp, high confidence
61	Qwen3.8-Flash-Next	2.4%	estimated ± 0.2 pp, high confidence
62	A.X K2	2.4%	estimated ± 0.2 pp, medium confidence
63	Apodex 1.1	2.4%	estimated ± 0.2 pp, medium confidence
64	Apodex 1.1 Mini	2.4%	estimated ± 0.2 pp, medium confidence
65	Celeris-1	2.4%	estimated ± 0.2 pp, medium confidence
66	Claude 3.5 Haiku	2.4%	estimated ± 0.2 pp, medium confidence
67	Claude 3.7 Sonnet	2.4%	estimated ± 0.2 pp, medium confidence
68	Claude 3.7 Sonnet Thinking	2.4%	estimated ± 0.2 pp, medium confidence
69	Claude 3 Haiku	2.4%	estimated ± 0.2 pp, medium confidence
70	Claude 4.1 Opus Thinking	2.4%	estimated ± 0.2 pp, medium confidence
71	Claude 4 Sonnet	2.4%	estimated ± 0.2 pp, medium confidence
72	Claude Opus 4.5	2.4%	estimated ± 0.2 pp, medium confidence
73	Claude Opus 4.5 Thinking	2.4%	estimated ± 0.2 pp, medium confidence
74	Claude Opus 4.6	2.4%	estimated ± 0.2 pp, medium confidence
75	Claude Opus 4.7	2.4%	estimated ± 0.2 pp, medium confidence
76	Claude Sonnet 4.6	2.4%	estimated ± 0.2 pp, medium confidence
77	Command A+	2.4%	estimated ± 0.2 pp, medium confidence
78	DeepSeek-R1	2.4%	estimated ± 0.2 pp, medium confidence
79	DeepSeek V3	2.4%	estimated ± 0.2 pp, medium confidence
80	DeepSeek V3 0324	2.4%	estimated ± 0.2 pp, medium confidence
81	DeepSeek V3.1	2.4%	estimated ± 0.2 pp, medium confidence
82	DeepSeek V3.1 (Reasoning)	2.4%	estimated ± 0.2 pp, medium confidence
83	DeepSeek V3.2	2.4%	estimated ± 0.2 pp, medium confidence
84	Devstral Small	2.4%	estimated ± 0.2 pp, medium confidence
85	Exaone 4.0 1.2B	2.4%	estimated ± 0.2 pp, medium confidence
86	Exaone 4.0 32B	2.4%	estimated ± 0.2 pp, medium confidence
87	Gemini 2.0 Flash	2.4%	estimated ± 0.2 pp, medium confidence
88	Gemini 2.5 Flash	2.4%	estimated ± 0.2 pp, medium confidence
89	Gemini 2.5 Flash-Lite	2.4%	estimated ± 0.2 pp, medium confidence
90	Gemini 2.5 Pro	2.4%	estimated ± 0.2 pp, medium confidence
91	Gemini 3.5 Flash-Lite	2.4%	estimated ± 0.2 pp, medium confidence
92	Gemini 3.6 Flash	2.4%	estimated ± 0.2 pp, high confidence
93	Gemini 3 Flash	2.4%	estimated ± 0.2 pp, medium confidence
94	Gemini 3 Pro	2.4%	estimated ± 0.2 pp, medium confidence
95	Gemma 3 27B	2.4%	estimated ± 0.2 pp, medium confidence
96	Gemma 4 12B	2.4%	estimated ± 0.2 pp, medium confidence
97	Gemma 4 26B A4B	2.4%	estimated ± 0.2 pp, medium confidence
98	Gemma 4 31B	2.4%	estimated ± 0.2 pp, medium confidence
99	Gemma 4 E2B	2.4%	estimated ± 0.2 pp, medium confidence
100	Gemma 4 E4B	2.4%	estimated ± 0.2 pp, medium confidence
101	GLM-4.5-Air	2.4%	estimated ± 0.2 pp, medium confidence
102	GLM-4.6	2.4%	estimated ± 0.2 pp, medium confidence
103	GLM-4.7	2.4%	estimated ± 0.2 pp, medium confidence
104	GLM-5	2.4%	estimated ± 0.2 pp, medium confidence
105	GLM-5.1	2.4%	estimated ± 0.2 pp, medium confidence
106	GLM-5-Turbo	2.4%	estimated ± 0.2 pp, medium confidence
107	GLM-5V-Turbo	2.4%	estimated ± 0.2 pp, medium confidence
108	GPT-4.1	2.4%	estimated ± 0.2 pp, medium confidence
109	GPT-4.1 mini	2.4%	estimated ± 0.2 pp, medium confidence
110	GPT-4.1 nano	2.4%	estimated ± 0.2 pp, medium confidence
111	GPT-4o	2.4%	estimated ± 0.2 pp, medium confidence
112	GPT-5.1	2.4%	estimated ± 0.2 pp, medium confidence
113	GPT-5.1-Codex	2.4%	estimated ± 0.2 pp, medium confidence
114	GPT-5.1-Codex-Max	2.4%	estimated ± 0.2 pp, medium confidence
115	GPT-5.2-Codex	2.4%	estimated ± 0.2 pp, medium confidence
116	GPT-5.4 mini	2.4%	estimated ± 0.2 pp, medium confidence
117	GPT-5.4 nano	2.4%	estimated ± 0.2 pp, medium confidence
118	GPT-5 (high)	2.4%	estimated ± 0.2 pp, medium confidence
119	GPT-5 (medium)	2.4%	estimated ± 0.2 pp, medium confidence
120	GPT-OSS 120B	2.4%	estimated ± 0.2 pp, medium confidence
121	GPT-OSS 20B	2.4%	estimated ± 0.2 pp, medium confidence
122	Granite-4.0-350M	2.4%	estimated ± 0.2 pp, medium confidence
123	Granite-4.0-H-1B	2.4%	estimated ± 0.2 pp, medium confidence
124	Granite-4.0-H-350M	2.4%	estimated ± 0.2 pp, medium confidence
125	Granite 4.2 30B	2.4%	estimated ± 0.2 pp, medium confidence
126	Granite 4.2 3B	2.4%	estimated ± 0.2 pp, medium confidence
127	Granite 4.2 8B	2.4%	estimated ± 0.2 pp, medium confidence
128	Grok 4	2.4%	estimated ± 0.2 pp, medium confidence
129	Grok 4.1 Fast	2.4%	estimated ± 0.2 pp, medium confidence
130	Grok 4.1 Fast (Reasoning)	2.4%	estimated ± 0.2 pp, medium confidence
131	Grok 4.3	2.4%	estimated ± 0.2 pp, medium confidence
132	Grok 4 Fast (Reasoning)	2.4%	estimated ± 0.2 pp, medium confidence
133	Grok Code Fast 1	2.4%	estimated ± 0.2 pp, medium confidence
134	Hy3	2.4%	estimated ± 0.2 pp, medium confidence
135	Hy3 Preview	2.4%	estimated ± 0.2 pp, medium confidence
136	Inkling	2.4%	estimated ± 0.2 pp, medium confidence
137	Inkling-Small	2.4%	estimated ± 0.2 pp, medium confidence
138	K-Exaone	2.4%	estimated ± 0.2 pp, medium confidence
139	K-EXAONE 2.0	2.4%	estimated ± 0.2 pp, medium confidence
140	Kimi K2.6	2.4%	estimated ± 0.2 pp, medium confidence
141	Kimi K2	2.4%	estimated ± 0.2 pp, medium confidence
142	Kimi K2.5	2.4%	estimated ± 0.2 pp, medium confidence
143	Kimi K2.5 (Reasoning)	2.4%	estimated ± 0.2 pp, medium confidence
144	Kimi K2.7 Code	2.4%	estimated ± 0.2 pp, medium confidence
145	LFM2.5-2.6B	2.4%	estimated ± 0.2 pp, medium confidence
146	LFM2.5-8B-A1B	2.4%	estimated ± 0.2 pp, medium confidence
147	LFM2.5-VL-1.6B-Extract	2.4%	estimated ± 0.2 pp, medium confidence
148	Ling 2.6 Flash	2.4%	estimated ± 0.2 pp, medium confidence
149	Ling 3.0 Flash	2.4%	estimated ± 0.2 pp, medium confidence
150	Ling 3.0 Flash VL	2.4%	estimated ± 0.2 pp, medium confidence
151	Ling 3.0 Tiny	2.4%	estimated ± 0.2 pp, medium confidence
152	Llama 3.1 405B	2.4%	estimated ± 0.2 pp, medium confidence
153	Llama 3.1 70B	2.4%	estimated ± 0.2 pp, medium confidence
154	Llama 3.3 70B	2.4%	estimated ± 0.2 pp, medium confidence
155	Llama 4 Maverick	2.4%	estimated ± 0.2 pp, medium confidence
156	Llama 4 Scout	2.4%	estimated ± 0.2 pp, medium confidence
157	Magistral Medium	2.4%	estimated ± 0.2 pp, medium confidence
158	Magistral Small	2.4%	estimated ± 0.2 pp, medium confidence
159	Mercury 2.5	2.4%	estimated ± 0.2 pp, medium confidence
160	MiMo-V2.5-Pro	2.4%	estimated ± 0.2 pp, medium confidence
161	MiMo-V2-Flash	2.4%	estimated ± 0.2 pp, medium confidence
162	MiMo-V2-Omni	2.4%	estimated ± 0.2 pp, medium confidence
163	MiMo-V2-Pro	2.4%	estimated ± 0.2 pp, medium confidence
164	MiniCPM5-2B	2.4%	estimated ± 0.2 pp, medium confidence
165	MiniMax M2.7	2.4%	estimated ± 0.2 pp, medium confidence
166	MiniMax M3	2.4%	estimated ± 0.2 pp, medium confidence
167	Mistral Large 2	2.4%	estimated ± 0.2 pp, medium confidence
168	Mistral Large 3	2.4%	estimated ± 0.2 pp, medium confidence
169	Mistral Medium 3	2.4%	estimated ± 0.2 pp, medium confidence
170	Mistral Medium 3.5 128B	2.4%	estimated ± 0.2 pp, medium confidence
171	Mistral Small 3.1	2.4%	estimated ± 0.2 pp, medium confidence
172	Mistral Small 4	2.4%	estimated ± 0.2 pp, medium confidence
173	Mistral Small 4 (Reasoning)	2.4%	estimated ± 0.2 pp, medium confidence
174	Muse Glimmer 30B	2.4%	estimated ± 0.2 pp, medium confidence
175	Nemotron 3.5 Bolt 74B-A9B	2.4%	estimated ± 0.2 pp, medium confidence
176	Nemotron 3.5 Lightning 30B A3B NVFP4	2.4%	estimated ± 0.2 pp, medium confidence
177	Nemotron 3.5 Super VL 120B A12B NVFP4	2.4%	estimated ± 0.2 pp, high confidence
178	Nemotron 3 Nano 30B	2.4%	estimated ± 0.2 pp, medium confidence
179	Nemotron 3 Nano Omni 30B A3B	2.4%	estimated ± 0.2 pp, medium confidence
180	Nemotron 3 Ultra	2.4%	estimated ± 0.2 pp, medium confidence
181	Nemotron Ultra 253B	2.4%	estimated ± 0.2 pp, medium confidence
182	North Mini Code	2.4%	estimated ± 0.2 pp, medium confidence
183	Nova Pro	2.4%	estimated ± 0.2 pp, medium confidence
184	o1	2.4%	estimated ± 0.2 pp, medium confidence
185	o3	2.4%	estimated ± 0.2 pp, medium confidence
186	o4-mini	2.4%	estimated ± 0.2 pp, medium confidence
187	Phi-4	2.4%	estimated ± 0.2 pp, medium confidence
188	Phi-4-reasoning	2.4%	estimated ± 0.2 pp, medium confidence
189	Quasar 438B	2.4%	estimated ± 0.2 pp, medium confidence
190	Qwen3.5-122B-A10B	2.4%	estimated ± 0.2 pp, medium confidence
191	Qwen3.5-27B	2.4%	estimated ± 0.2 pp, medium confidence
192	Qwen3.5-35B-A3B	2.4%	estimated ± 0.2 pp, medium confidence
193	Qwen3.5 397B	2.4%	estimated ± 0.2 pp, medium confidence
194	Qwen3.5 397B (Reasoning)	2.4%	estimated ± 0.2 pp, medium confidence
195	Qwen3.5-4B	2.4%	estimated ± 0.2 pp, medium confidence
196	Qwen3.6-27B	2.4%	estimated ± 0.2 pp, medium confidence
197	Qwen3.6-35B-A3B	2.4%	estimated ± 0.2 pp, medium confidence
198	Qwen 3.6 Max (preview)	2.4%	estimated ± 0.2 pp, medium confidence
199	Qwen3.6 Plus	2.4%	estimated ± 0.2 pp, medium confidence
200	Qwen3.7 Plus	2.4%	estimated ± 0.2 pp, medium confidence
201	Qwen3.8-27B	2.4%	estimated ± 0.2 pp, medium confidence
202	Qwen3 Max	2.4%	estimated ± 0.2 pp, medium confidence
203	Qwen3-Omni-30B-A3B-Instruct	2.4%	estimated ± 0.2 pp, medium confidence
204	Qwen3-Omni-30B-A3B-Thinking	2.4%	estimated ± 0.2 pp, medium confidence
205	Sarvam 105B	2.4%	estimated ± 0.2 pp, medium confidence
206	Sarvam 30B	2.4%	estimated ± 0.2 pp, medium confidence
207	Solar Pro 2	2.4%	estimated ± 0.2 pp, medium confidence
208	Solar Pro 3	2.4%	estimated ± 0.2 pp, medium confidence
209	Solar Pro 4	2.4%	estimated ± 0.2 pp, medium confidence
210	Step 3.7 Flash	2.4%	estimated ± 0.2 pp, medium confidence
211	Trinity-Large-Preview	2.4%	estimated ± 0.2 pp, medium confidence
212	Trinity-Large-Thinking	2.4%	estimated ± 0.2 pp, medium confidence
213	Ultravox v0.6 Llama 3.3 70B	2.4%	estimated ± 0.2 pp, medium confidence
214	Gemini 4 Argon	2.4%	measured
215	Nemotron 3.5 Bolt 74B-A9B BF16	2.4%	estimated ± 0.2 pp, low confidence
216	LFM3-1.7B	2.2%	estimated ± 0.2 pp, low confidence
217	MiMo-V2.6-Pro	2.2%	measured

STEMBench leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General