Vision & documents

MMMU-Pro leaderboard

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

As of 2026-10-07, the highest measured score on MMMU-Pro is 83.9% by Gemini 3.1 Pro. 88 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Gemini 3.1 Pro	83.9%	measured
2	Gemini 3.5 Flash	83.6%	measured
3	GPT-5.6 Sol	83.0%	measured
4	Claude Opus 5.5	82.5%	estimated ± 4.4 pp, low confidence
5	Qwen3.8 Max	82.3%	measured
6	Claude Mythos 5	82.3%	estimated ± 1.4 pp, low confidence
7	Gemini 3.8 Flash	82.2%	estimated ± 1.4 pp, medium confidence
8	GPT-6 Astra	81.9%	estimated ± 2.7 pp, medium confidence
9	GPT-6.1 Sol	81.8%	estimated ± 4.4 pp, low confidence
10	Step 3.7 Flash	81.8%	estimated ± 2.4 pp, high confidence
11	Kimi K3	81.6%	measured
12	Seed 2.1 Pro	81.6%	measured
13	Qwen3.8-Flash-Next	81.5%	estimated ± 1.4 pp, medium confidence
14	GLM-5.3-Flash	81.4%	estimated ± 3.2 pp, high confidence
15	Gemini 3.7 Flash	81.4%	estimated ± 1.4 pp, medium confidence
16	Claude Opus 5	81.3%	estimated ± 4.4 pp, low confidence
17	GPT-5.4	81.2%	measured
18	GPT-5.5	81.2%	measured
19	Muse Spark 1.1	81.1%	estimated ± 3.2 pp, high confidence
20	Gemini 3 Pro	81.0%	measured
21	GPT-5.6 Terra	80.7%	measured
22	Gemini 3.6 Flash	80.6%	estimated ± 4.4 pp, medium confidence
23	GPT-6 Sol	80.5%	estimated ± 4.4 pp, medium confidence
24	Sakana Fugu-Ultra	80.5%	estimated ± 3.2 pp, high confidence
25	Qwen3.8 Max Preview	80.4%	estimated ± 4.4 pp, medium confidence
26	Muse Spark	80.4%	measured
27	Qwen3.8-27B	80.3%	estimated ± 1.4 pp, medium confidence
28	Seed 2.1 Turbo	80.1%	measured
29	Qwen3.8-Omni-Flash	79.9%	estimated ± 1.4 pp, medium confidence
30	Sakana Fugu	79.9%	estimated ± 3.2 pp, high confidence
31	GPT-5.2	79.5%	measured
32	Kimi K2.6	79.4%	measured
33	Grok 4.5	79.4%	estimated ± 4.4 pp, medium confidence
34	dots3-note Preview	79.1%	measured
35	GPT-6 Luna	79.1%	estimated ± 4.4 pp, medium confidence
36	Qwen3.5 397B	79.0%	measured
37	Qwen3.7 Plus	79.0%	measured
38	Apodex 1.1	78.8%	estimated ± 4.4 pp, medium confidence
39	Apodex 1.1 Mini	78.8%	estimated ± 4.4 pp, medium confidence
40	Qwen3.6 Plus	78.8%	measured
41	Gemini 3.5 Flash-Lite	78.8%	estimated ± 4.4 pp, medium confidence
42	Ling 3.0 Flash VL	78.8%	estimated ± 4.4 pp, medium confidence
43	Gemini 3 Flash	78.6%	estimated ± 4.4 pp, medium confidence
44	GPT-5.3 Codex	78.5%	estimated ± 4.4 pp, medium confidence
45	Kimi K2.5	78.5%	measured
46	Kimi K2.5 (Reasoning)	78.5%	measured
47	GPT-5.6 Luna	78.4%	measured
48	Holo2-235B-A22B	78.3%	estimated ± 2.7 pp, high confidence
49	Grok 4.3	78.1%	measured
50	MiniMax M3	78.1%	measured
51	Pareto 26.9	78.0%	measured
52	Qwen3.5-122B-A10B	77.9%	estimated ± 2.2 pp, high confidence
53	MiMo-V2.5	77.9%	measured
54	DeepSeek V4.1 Flash	77.9%	estimated ± 4.4 pp, medium confidence
55	Qwen3.5-27B	77.8%	estimated ± 2.2 pp, high confidence
56	Claude Opus 4.7	77.6%	estimated ± 4.4 pp, medium confidence
57	Mistral Large 4	77.6%	estimated ± 4.4 pp, medium confidence
58	GPT-5.2-Codex	77.5%	estimated ± 4.4 pp, medium confidence
59	Holo2-30B-A3B	77.3%	estimated ± 2.7 pp, high confidence
60	Claude Opus 4.6	77.3%	measured
61	GPT-5.1	77.2%	estimated ± 4.4 pp, medium confidence
62	Claude Opus 4.6 (Adaptive)	77.1%	estimated ± 4.4 pp, medium confidence
63	Claude Sonnet 4.6	77.1%	estimated ± 3.2 pp, high confidence
64	Gemini 2.5 Pro	76.9%	estimated ± 4.4 pp, medium confidence
65	Gemma 4 31B	76.9%	measured
66	Claude Opus 4.7 (Adaptive)	76.7%	estimated ± 1.4 pp, medium confidence
67	GPT-5 (medium)	76.6%	estimated ± 4.4 pp, medium confidence
68	Qwen3.5-35B-A3B	76.6%	estimated ± 2.2 pp, high confidence
69	GPT-5.4 mini	76.6%	measured
70	GPT-5 (high)	76.6%	estimated ± 4.4 pp, medium confidence
71	Claude Opus 4.5 Thinking	76.5%	estimated ± 4.4 pp, medium confidence
72	MiMo-V2.6-Flash	76.1%	estimated ± 4.4 pp, medium confidence
73	Step 5 Preview	76.0%	measured
74	GLM-5V-Turbo	75.9%	estimated ± 4.4 pp, medium confidence
75	Qwen3.6-27B	75.8%	measured
76	GPT-5.1-Codex	75.8%	estimated ± 4.4 pp, medium confidence
77	GPT-5.1-Codex-Max	75.8%	estimated ± 4.4 pp, medium confidence
78	Holo2-8B	75.5%	estimated ± 2.7 pp, high confidence
79	Gemini 3.1 Flash-Lite	75.4%	estimated ± 3.2 pp, high confidence
80	Qwen3.6-35B-A3B	75.3%	measured
81	Grok 4.20	75.2%	measured
82	Holo2-4B	75.0%	estimated ± 2.7 pp, high confidence
83	o3	74.7%	estimated ± 4.4 pp, medium confidence
84	MiMo-V2-Omni	74.6%	estimated ± 4.4 pp, medium confidence
85	Claude Opus 4.8	74.4%	estimated ± 1.4 pp, medium confidence
86	Grok 4	74.1%	estimated ± 4.4 pp, medium confidence
87	Inkling-Small	74.0%	measured
88	Muse Glimmer 30B	74.0%	measured
89	Gemma 4 26B A4B	73.8%	measured
90	Claude Sonnet 5	73.7%	estimated ± 1.4 pp, low confidence
91	Claude 4.1 Opus Thinking	73.6%	estimated ± 4.4 pp, medium confidence
92	Inkling	73.5%	measured
93	Gemini 2.5 Flash	72.5%	estimated ± 4.4 pp, medium confidence
94	Mistral Medium 3.5 128B	72.2%	estimated ± 4.4 pp, medium confidence
95	Grok 4.1 Fast (Reasoning)	71.4%	estimated ± 4.4 pp, medium confidence
96	Interfaze Beta	71.1%	measured
97	Claude 4 Sonnet	71.0%	estimated ± 4.4 pp, medium confidence
98	Llama 4 Maverick	70.8%	estimated ± 4.4 pp, medium confidence
99	Grok 4 Fast (Reasoning)	70.7%	estimated ± 4.4 pp, medium confidence
100	Claude Opus 4.5	70.6%	measured
101	GPT-4.1	70.4%	estimated ± 4.4 pp, medium confidence
102	Qwen3-Omni-30B-A3B-Thinking	69.9%	estimated ± 4.4 pp, medium confidence
103	GPT-4.1 mini	69.1%	estimated ± 4.4 pp, medium confidence
104	Gemma 4 12B	69.1%	measured
105	Mistral Small 4	68.2%	estimated ± 4.4 pp, medium confidence
106	Mistral Small 4 (Reasoning)	68.2%	estimated ± 4.4 pp, medium confidence
107	Mistral Large 3	67.6%	estimated ± 4.4 pp, medium confidence
108	Qwen3-Omni-30B-A3B-Instruct	67.5%	estimated ± 4.4 pp, medium confidence
109	Gemini 1.5 Pro	67.2%	estimated ± 4.4 pp, medium confidence
110	Mistral Medium 3	66.2%	estimated ± 4.4 pp, medium confidence
111	Llama 4 Scout	66.1%	estimated ± 4.4 pp, medium confidence
112	GPT-5.4 nano	66.1%	measured
113	Qwen3.5 397B (Reasoning)	66.0%	estimated ± 4.4 pp, medium confidence
114	Gemma 4 E4B	65.4%	estimated ± 4.4 pp, low confidence
115	Grok 4.1 Fast	63.8%	estimated ± 4.4 pp, low confidence
116	Gemma 3 27B	63.5%	estimated ± 4.4 pp, low confidence
117	Command A+	63.0%	measured
118	Ternary Bonsai 2 27B	61.8%	estimated ± 11.4 pp, low confidence
119	Gemma 4 E2B	61.7%	estimated ± 4.4 pp, low confidence
120	Nova Pro	61.5%	estimated ± 4.4 pp, low confidence
121	GPT-4o mini	60.0%	estimated ± 4.4 pp, low confidence
122	GPT-4.1 nano	59.2%	estimated ± 4.4 pp, low confidence
123	Nemotron 3 Nano Omni 30B A3B	57.2%	estimated ± 2.4 pp, medium confidence
124	Claude 3 Haiku	53.7%	estimated ± 4.4 pp, low confidence
125	LFM2.5-VL-1.6B-Extract	51.1%	estimated ± 4.4 pp, low confidence
126	Phi-4 Multimodal Instruct	43.3%	estimated ± 4.4 pp, low confidence
127	LFM2.5-VL-3B	30.5%	measured
128	ZAYA1-VL-8B	28.2%	estimated ± 2.4 pp, low confidence
129	North Micro Vision Instruct	22.6%	estimated ± 11.4 pp, low confidence
130	LFM2.5-VL-450M	16.0%	estimated ± 2.4 pp, low confidence

MMMU-Pro leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General