Knowledge & reasoning

MMLU-Pro leaderboard

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

As of 2026-10-07, the highest measured score on MMLU-Pro is 89.6% by Qwen3.7 Max. 190 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Qwen3.7 Max	89.6%	measured
2	Claude Opus 4.5	89.5%	measured
3	Qwen3.6 Plus	88.5%	measured
4	Qwen3.7 Plus	88.5%	measured
5	Pareto 26.10 Preview	88.5%	estimated ± 5.4 pp, medium confidence
6	Gemini 3 Pro Deep Think	88.5%	estimated ± 5.8 pp, low confidence
7	GLM-5.3-Flash	88.2%	estimated ± 3.9 pp, high confidence
8	Qwen3.5 397B	87.8%	measured
9	GPT-6 Astra	87.6%	estimated ± 3.0 pp, medium confidence
10	DeepSeek V4 Pro 0813	87.5%	measured
11	GPT-5.6 Sol	87.4%	estimated ± 3.0 pp, medium confidence
12	Kimi K2.5	87.1%	measured
13	Kimi K2.5 (Reasoning)	87.1%	measured
14	GPT-5.6 Terra	87.0%	estimated ± 3.0 pp, medium confidence
15	GPT-5.2	86.9%	estimated ± 3.0 pp, high confidence
16	GPT-5.6 Luna	86.9%	estimated ± 3.0 pp, high confidence
17	Gemini 3.5 Flash	86.8%	estimated ± 3.0 pp, high confidence
18	Nemotron 3 Ultra	86.8%	measured
19	Qwen3.5-122B-A10B	86.7%	measured
20	Qwen3.8-Omni-Flash	86.6%	estimated ± 3.0 pp, high confidence
21	DeepSeek V4.1 Flash	86.5%	estimated ± 3.0 pp, high confidence
22	Kimi K2.6	86.5%	estimated ± 3.0 pp, high confidence
23	Grok 4.3	86.4%	estimated ± 3.0 pp, high confidence
24	dots3-note Preview	86.3%	estimated ± 3.6 pp, high confidence
25	Agents-A1	86.3%	estimated ± 3.6 pp, high confidence
26	Interfaze Beta	86.3%	estimated ± 3.0 pp, high confidence
27	Solar Pro 4	86.3%	measured
28	DeepSeek V4 Flash 0731	86.2%	measured
29	Qwen3.6-27B	86.2%	measured
30	Solar Open 2	86.2%	measured
31	Qwen3.5-27B	86.1%	measured
32	Claude Opus 5.5	86.1%	estimated ± 2.5 pp, low confidence
33	Claude Fable 5.1	86.0%	estimated ± 2.5 pp, low confidence
34	Claude Mythos 5	86.0%	estimated ± 2.5 pp, low confidence
35	Claude Sonnet 5.5	86.0%	estimated ± 2.5 pp, low confidence
36	Claude Opus 5	86.0%	estimated ± 2.5 pp, low confidence
37	Muse Spark 1.1	85.9%	estimated ± 2.5 pp, low confidence
38	Sakana Fugu-Ultra	85.8%	estimated ± 2.5 pp, low confidence
39	Claude Opus 4.8	85.8%	estimated ± 2.5 pp, low confidence
40	Pareto 26.9	85.8%	estimated ± 2.5 pp, low confidence
41	Sakana Fugu	85.8%	estimated ± 2.5 pp, low confidence
42	Claude Opus 4.7 (Adaptive)	85.8%	estimated ± 2.5 pp, low confidence
43	Claude Haiku 5.5	85.8%	estimated ± 2.5 pp, low confidence
44	Claude Fable 5	85.7%	estimated ± 3.5 pp, medium confidence
45	GPT-6.1 Sol	85.7%	estimated ± 3.5 pp, medium confidence
46	Gemini 3.7 Flash	85.7%	estimated ± 3.5 pp, medium confidence
47	Gemini 3.8 Flash	85.7%	estimated ± 3.5 pp, medium confidence
48	Gemini 3 Pro	85.7%	estimated ± 3.5 pp, medium confidence
49	GPT-5.3 Codex	85.7%	estimated ± 3.5 pp, medium confidence
50	GPT-6 Sol	85.7%	estimated ± 3.5 pp, medium confidence
51	Grok 4.5	85.7%	estimated ± 3.5 pp, medium confidence
52	Gemini 3.6 Flash	85.7%	estimated ± 3.5 pp, medium confidence
53	Gemini 4 Argon	85.7%	estimated ± 3.5 pp, medium confidence
54	Grok 4.6	85.7%	estimated ± 3.5 pp, high confidence
55	Claude Opus 4.6 (Adaptive)	85.7%	estimated ± 3.5 pp, high confidence
56	Grok 4.7	85.7%	estimated ± 3.5 pp, high confidence
57	Claude Opus 4.5 Thinking	85.7%	estimated ± 3.5 pp, high confidence
58	Gemini 3 Flash	85.7%	estimated ± 3.5 pp, high confidence
59	Muse Spark 1.2	85.7%	estimated ± 3.5 pp, high confidence
60	Claude Opus 4.7	85.7%	estimated ± 3.5 pp, high confidence
61	GPT-6 Luna	85.7%	estimated ± 3.5 pp, high confidence
62	Muse Spark 1.3	85.7%	estimated ± 3.5 pp, high confidence
63	Step 5 Preview	85.7%	estimated ± 3.5 pp, high confidence
64	GPT-5.2-Codex	85.7%	estimated ± 3.5 pp, high confidence
65	Grok 4	85.7%	estimated ± 3.5 pp, high confidence
66	GPT-5 (high)	85.7%	estimated ± 3.5 pp, high confidence
67	GPT-5.1-Codex	85.7%	estimated ± 3.5 pp, high confidence
68	GPT-5.1-Codex-Max	85.7%	estimated ± 3.5 pp, high confidence
69	Kimi K2.7 Code	85.7%	estimated ± 3.5 pp, high confidence
70	GPT-5 (medium)	85.7%	estimated ± 3.5 pp, high confidence
71	Gemini 3.1 Pro	85.7%	estimated ± 2.5 pp, low confidence
72	o3	85.7%	estimated ± 3.5 pp, high confidence
73	Qwen 3.6 Max (preview)	85.7%	estimated ± 3.5 pp, high confidence
74	GPT-5.1	85.7%	estimated ± 3.5 pp, high confidence
75	MiMo-V2.6-Pro	85.7%	estimated ± 3.5 pp, high confidence
76	GLM-5.3	85.7%	estimated ± 3.5 pp, high confidence
77	Ornith-1.5-397B	85.7%	estimated ± 2.5 pp, low confidence
78	Hy3	85.7%	estimated ± 3.5 pp, high confidence
79	Apodex 1.1	85.7%	estimated ± 3.5 pp, high confidence
80	Apodex 1.1 Mini	85.7%	estimated ± 3.5 pp, high confidence
81	Qwen3.8 Max Preview	85.7%	estimated ± 3.5 pp, high confidence
82	DeepSeek-R1	85.7%	estimated ± 3.5 pp, high confidence
83	GLM-5	85.7%	measured
84	Qwen3.8 Max	85.7%	estimated ± 2.5 pp, low confidence
85	Kimi K3	85.7%	estimated ± 2.5 pp, low confidence
86	Gemini 3.5 Flash-Lite	85.7%	estimated ± 3.5 pp, high confidence
87	Hy4 preview	85.7%	estimated ± 2.5 pp, low confidence
88	GLM-5V-Turbo	85.7%	estimated ± 3.5 pp, high confidence
89	Claude Sonnet 5	85.7%	estimated ± 2.5 pp, low confidence
90	Ling 3.1 Flash	85.7%	estimated ± 3.5 pp, high confidence
91	DeepSeek V3.1 (Reasoning)	85.7%	estimated ± 3.5 pp, high confidence
92	GPT-5.5 Pro	85.7%	estimated ± 2.5 pp, low confidence
93	Muse Spark	85.7%	estimated ± 2.5 pp, low confidence
94	GPT-5.4 Pro	85.7%	estimated ± 2.5 pp, low confidence
95	GLM-5-Turbo	85.7%	estimated ± 3.5 pp, high confidence
96	Kimi K2	85.6%	estimated ± 3.5 pp, high confidence
97	MiMo-V2.6-Flash	85.6%	estimated ± 3.5 pp, high confidence
98	Muse Glimmer 30B	85.6%	estimated ± 3.5 pp, high confidence
99	GPT-5.5	85.6%	estimated ± 2.5 pp, low confidence
100	MiniMax M2.7	85.6%	estimated ± 3.5 pp, high confidence
101	MiMo-V2-Pro	85.6%	estimated ± 3.5 pp, high confidence
102	Hy3 Preview	85.6%	estimated ± 3.0 pp, high confidence
103	GLM-5.2	85.6%	estimated ± 2.5 pp, low confidence
104	Gemini 2.5 Flash	85.6%	estimated ± 3.5 pp, high confidence
105	Mistral Large 4	85.6%	estimated ± 3.5 pp, high confidence
106	Step 3.7 Flash	85.6%	estimated ± 3.5 pp, high confidence
107	GPT-5.4	85.6%	estimated ± 2.5 pp, medium confidence
108	Grok 4.1 Fast (Reasoning)	85.6%	estimated ± 3.5 pp, high confidence
109	Seed 2.1 Pro	85.6%	estimated ± 6.8 pp, medium confidence
110	Mistral Large 3	85.6%	estimated ± 3.5 pp, high confidence
111	Llama 4 Maverick	85.5%	estimated ± 3.5 pp, high confidence
112	Mistral Medium 3.5 128B	85.5%	estimated ± 3.5 pp, high confidence
113	Qwen3.5 397B (Reasoning)	85.5%	estimated ± 3.5 pp, high confidence
114	Qwen3 Max	85.5%	estimated ± 3.5 pp, high confidence
115	DeepSeek V3 0324	85.5%	estimated ± 3.5 pp, high confidence
116	Nemotron 3 Super 100B	85.5%	estimated ± 3.5 pp, high confidence
117	DeepSeek V3.2	85.5%	estimated ± 3.5 pp, high confidence
118	GLM-5.1	85.5%	estimated ± 3.5 pp, high confidence
119	Beam	85.5%	estimated ± 2.5 pp, medium confidence
120	Qwen3.8-Flash-Next	85.5%	estimated ± 2.5 pp, medium confidence
121	Grok Code Fast 1	85.5%	estimated ± 3.5 pp, high confidence
122	Llama 3.1 405B	85.4%	estimated ± 3.5 pp, high confidence
123	DeepSeek V3.1	85.4%	estimated ± 3.5 pp, high confidence
124	Grok 4 Fast (Reasoning)	85.4%	estimated ± 3.5 pp, high confidence
125	Claude 4 Sonnet	85.4%	estimated ± 3.5 pp, high confidence
126	MiMo-V2.5-Pro	85.4%	estimated ± 2.5 pp, medium confidence
127	Trinity-Large-Preview	85.4%	estimated ± 3.5 pp, high confidence
128	Trinity-Large-Thinking	85.4%	estimated ± 3.5 pp, high confidence
129	Qwen3.5-35B-A3B	85.3%	measured
130	Mercury 2.5	85.3%	estimated ± 3.5 pp, high confidence
131	GPT-OSS 120B	85.3%	estimated ± 3.5 pp, high confidence
132	Grok 4.20	85.3%	estimated ± 2.5 pp, medium confidence
133	Inkling-Small	85.3%	estimated ± 2.5 pp, medium confidence
134	Mistral Small 4	85.3%	estimated ± 3.5 pp, high confidence
135	Mistral Small 4 (Reasoning)	85.3%	estimated ± 3.5 pp, high confidence
136	o3-pro	85.2%	estimated ± 3.9 pp, high confidence
137	Qwen3.8-27B	85.2%	estimated ± 2.5 pp, medium confidence
138	GLM-4.6	85.2%	estimated ± 3.5 pp, high confidence
139	Gemma 4 31B	85.2%	measured
140	Qwen3.6-35B-A3B	85.2%	measured
141	Inkling	85.2%	estimated ± 2.5 pp, medium confidence
142	GPT-5.4 mini	85.1%	estimated ± 2.5 pp, medium confidence
143	MAI-Thinking-1	85.0%	measured
144	Nemotron Ultra 253B	85.0%	estimated ± 3.5 pp, high confidence
145	Ling 3.0 Flash	85.0%	estimated ± 3.0 pp, high confidence
146	GPT-4o	84.9%	estimated ± 3.5 pp, high confidence
147	Mistral Large 2	84.9%	estimated ± 3.5 pp, high confidence
148	Ornith-1.5-35B-A3B	84.9%	estimated ± 2.5 pp, medium confidence
149	MiMo-V2-Flash	84.9%	measured
150	GPT-5.4 nano	84.8%	estimated ± 2.5 pp, medium confidence
151	MiMo-V2-Omni	84.8%	estimated ± 3.5 pp, high confidence
152	Claude 4.1 Opus	84.8%	estimated ± 4.5 pp, high confidence
153	Ultravox v0.6 Llama 3.3 70B	84.7%	estimated ± 3.5 pp, high confidence
154	Ling 3.0 Flash FP8	84.7%	estimated ± 3.0 pp, high confidence
155	North Mini Code	84.7%	estimated ± 3.5 pp, high confidence
156	A.X K2	84.6%	estimated ± 3.5 pp, high confidence
157	Solar Pro 3	84.6%	estimated ± 3.5 pp, high confidence
158	Claude Sonnet 4.5	84.5%	estimated ± 3.0 pp, high confidence
159	Mistral Medium 3	84.5%	estimated ± 3.5 pp, high confidence
160	Seed 2.1 Turbo	84.5%	estimated ± 6.8 pp, medium confidence
161	Ornith-1.5-9B	84.4%	estimated ± 2.5 pp, medium confidence
162	Gemini 2.5 Pro	84.4%	estimated ± 3.0 pp, high confidence
163	GLM-4.7	84.3%	measured
164	Claude 3 Haiku	84.2%	estimated ± 3.5 pp, high confidence
165	Sarvam 105B	84.2%	estimated ± 3.5 pp, high confidence
166	Nemotron 3 Nano 30B	84.1%	estimated ± 3.5 pp, high confidence
167	Grok 4.1 Fast	84.0%	estimated ± 3.5 pp, high confidence
168	Nova Pro	83.9%	estimated ± 3.5 pp, high confidence
169	MiniMax M3	83.8%	estimated ± 3.5 pp, high confidence
170	Claude 4.1 Opus Thinking	83.7%	estimated ± 3.9 pp, high confidence
171	K-Exaone	83.6%	estimated ± 3.5 pp, high confidence
172	GLM-4.5-Air	83.6%	estimated ± 3.5 pp, high confidence
173	K-EXAONE 2.0	83.5%	measured
174	Solar Pro 2	83.4%	estimated ± 3.5 pp, high confidence
175	GPT-OSS 20B	83.4%	estimated ± 3.5 pp, high confidence
176	Quasar 438B	83.0%	estimated ± 3.5 pp, high confidence
177	Qwen3 235B 2507	83.0%	measured
178	o1-pro	83.0%	estimated ± 3.0 pp, high confidence
179	Llama 4 Scout	82.8%	estimated ± 3.5 pp, high confidence
180	Gemma 4 26B A4B	82.6%	measured
181	o3-mini	82.3%	estimated ± 3.0 pp, high confidence
182	Qwen3-Omni-30B-A3B-Thinking	82.3%	estimated ± 3.5 pp, high confidence
183	Ling 3.0 Flash VL	82.1%	estimated ± 3.5 pp, high confidence
184	Claude Opus 4.6	82.0%	measured
185	Qwen3-Omni-30B-A3B-Instruct	82.0%	estimated ± 3.5 pp, high confidence
186	Exaone 4.0 32B	81.8%	measured
187	Phi-4	81.8%	estimated ± 3.5 pp, high confidence
188	o1-preview	81.7%	estimated ± 3.9 pp, high confidence
189	o1	81.7%	estimated ± 3.0 pp, high confidence
190	Nemotron 3.5 Lightning 30B A3B NVFP4	81.6%	measured
191	Gemma 3 27B	80.4%	estimated ± 3.5 pp, high confidence
192	Sarvam 30B	79.8%	estimated ± 3.5 pp, high confidence
193	LongCat-Flash-Lite-Sparse	79.2%	measured
194	Claude Sonnet 4.6	79.2%	measured
195	Ternary Bonsai 2 27B	77.9%	estimated ± 1.4 pp, high confidence
196	Granite 4.2 30B	77.6%	measured
197	Nemotron 3 Nano Omni 30B A3B	77.3%	measured
198	Gemma 4 12B	77.2%	measured
199	GPT-4.1	76.8%	estimated ± 3.0 pp, high confidence
200	Celeris-1	75.9%	measured
201	DeepSeek V3	75.9%	measured
202	GPT-4.1 mini	75.5%	estimated ± 3.0 pp, high confidence
203	DeepSeek R1 Distill Qwen 32B	75.2%	estimated ± 3.9 pp, high confidence
204	ZAYA1-8B	74.2%	measured
205	Granite 4.2 8B	74.0%	measured
206	Gemini 1.5 Pro	74.0%	estimated ± 3.9 pp, high confidence
207	Mellum2-12B-A2.5B-Thinking	73.0%	estimated ± 1.4 pp, high confidence
208	LFM2.5-8B-A1B	72.4%	estimated ± 3.5 pp, high confidence
209	Claude 3.5 Sonnet	71.8%	estimated ± 3.0 pp, high confidence
210	GPT-4 Turbo	71.8%	estimated ± 4.5 pp, high confidence
211	Ling 2.6 Flash	71.5%	estimated ± 3.0 pp, high confidence
212	MiniCPM5-2B	70.8%	measured
213	Command A+	70.8%	estimated ± 3.5 pp, high confidence
214	Claude 3 Opus	69.6%	estimated ± 3.9 pp, high confidence
215	Gemma 4 E4B	69.4%	measured
216	Ling 3.0 Tiny	69.3%	estimated ± 3.5 pp, high confidence
217	ZAYA1-74B-Preview	68.1%	measured
218	Granite 4.2 3B	67.8%	measured
219	GPT-4o mini	66.9%	estimated ± 3.9 pp, medium confidence
220	Qwen2.5 Coder 32B Instruct	66.5%	estimated ± 3.9 pp, medium confidence
221	GPT-4.1 nano	62.7%	estimated ± 3.0 pp, high confidence
222	Phi-4 Multimodal Instruct	62.0%	estimated ± 3.9 pp, medium confidence
223	Mellum2-12B-A2.5B-Instruct	60.4%	estimated ± 1.4 pp, high confidence
224	Gemini 1.0 Pro	60.3%	estimated ± 3.9 pp, medium confidence
225	Gemma 4 E2B	60.0%	measured
226	LFM2.5-VL-1.6B-Extract	56.9%	estimated ± 3.5 pp, medium confidence
227	LLaDA2.2-mini	55.2%	estimated ± 5.4 pp, medium confidence
228	Granite-4.0-H-1B	53.5%	estimated ± 3.5 pp, medium confidence
229	Exaone 4.0 1.2B	52.4%	estimated ± 3.5 pp, medium confidence
230	Soofi S 30B-A3B	51.4%	measured
231	LFM2.5-2.6B	48.9%	estimated ± 3.5 pp, medium confidence
232	MiniCPM5-1B	48.9%	measured
233	Granite-4.0-350M	45.9%	estimated ± 3.5 pp, medium confidence
234	Granite-4.0-H-350M	45.3%	estimated ± 3.5 pp, medium confidence
235	LFM2.5-230M	20.3%	measured
236	LFM2.5-VL-450M	19.3%	measured

MMLU-Pro leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General