Vision & documents

MMMU leaderboard

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

As of 2026-10-07, the highest measured score on MMMU is 86.0% by Qwen3.6 Plus. 108 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Opus 5.5	94.0%	estimated ± 0.9 pp, medium confidence
2	GPT-6 Astra	93.3%	estimated ± 0.9 pp, medium confidence
3	GPT-6.1 Sol	92.5%	estimated ± 0.9 pp, medium confidence
4	Gemini 3.8 Flash	92.2%	estimated ± 0.9 pp, medium confidence
5	Gemini 3.7 Flash	92.1%	estimated ± 0.9 pp, medium confidence
6	Qwen3.8-Omni-Flash	91.7%	estimated ± 6.6 pp, low confidence
7	Claude Opus 5	91.4%	estimated ± 0.9 pp, medium confidence
8	Gemini 3.5 Flash	91.0%	estimated ± 0.9 pp, medium confidence
9	GPT-5.6 Sol	90.3%	estimated ± 0.9 pp, medium confidence
10	Gemini 3.6 Flash	90.1%	estimated ± 0.9 pp, medium confidence
11	GPT-6 Sol	89.8%	estimated ± 0.9 pp, medium confidence
12	Qwen3.8 Max Preview	89.7%	estimated ± 0.9 pp, medium confidence
13	Gemini 3.1 Pro	89.4%	estimated ± 0.9 pp, medium confidence
14	GPT-5.6 Terra	87.9%	estimated ± 0.9 pp, medium confidence
15	Kimi K3	87.7%	estimated ± 0.9 pp, medium confidence
16	Muse Spark	87.7%	estimated ± 0.9 pp, medium confidence
17	Qwen3.7 Plus	87.7%	estimated ± 0.9 pp, medium confidence
18	Grok 4.5	87.6%	estimated ± 0.9 pp, medium confidence
19	Gemini 3 Pro	87.4%	estimated ± 0.9 pp, medium confidence
20	GPT-5.5	87.2%	estimated ± 0.9 pp, medium confidence
21	Qwen3.8-Flash-Next	87.1%	estimated ± 0.9 pp, medium confidence
22	GPT-6 Luna	87.0%	estimated ± 0.9 pp, medium confidence
23	Qwen3.8 Max	86.9%	estimated ± 1.1 pp, low confidence
24	Kimi K2.6	86.7%	estimated ± 0.9 pp, medium confidence
25	Apodex 1.1	86.6%	estimated ± 0.9 pp, medium confidence
26	Apodex 1.1 Mini	86.6%	estimated ± 0.9 pp, medium confidence
27	Seed 2.1 Pro	86.5%	estimated ± 1.1 pp, low confidence
28	Gemini 3.5 Flash-Lite	86.4%	estimated ± 0.9 pp, medium confidence
29	Ling 3.0 Flash VL	86.4%	estimated ± 0.9 pp, medium confidence
30	Claude Opus 4.7 (Adaptive)	86.2%	estimated ± 0.9 pp, medium confidence
31	Gemini 3 Flash	86.0%	estimated ± 0.9 pp, medium confidence
32	GPT-5.6 Luna	86.0%	estimated ± 0.9 pp, medium confidence
33	MiniMax M3	86.0%	estimated ± 0.9 pp, medium confidence
34	Qwen3.6 Plus	86.0%	measured
35	GPT-5.3 Codex	85.9%	estimated ± 0.9 pp, medium confidence
36	GPT-5.4	85.9%	estimated ± 0.9 pp, medium confidence
37	Seed 2.1 Turbo	85.7%	estimated ± 1.1 pp, low confidence
38	Grok 4.3	85.6%	estimated ± 0.9 pp, medium confidence
39	GPT-5.2	85.3%	estimated ± 1.1 pp, low confidence
40	dots3-note Preview	85.1%	estimated ± 1.1 pp, low confidence
41	Claude Sonnet 5	84.9%	estimated ± 0.9 pp, high confidence
42	DeepSeek V4.1 Flash	84.7%	estimated ± 0.9 pp, high confidence
43	Pareto 26.9	84.4%	estimated ± 1.1 pp, medium confidence
44	MiMo-V2.5	84.4%	estimated ± 1.1 pp, medium confidence
45	Claude Opus 4.7	84.2%	estimated ± 0.9 pp, high confidence
46	Mistral Large 4	84.2%	estimated ± 0.9 pp, high confidence
47	Step 5 Preview	84.2%	estimated ± 0.9 pp, high confidence
48	GPT-5.2-Codex	84.1%	estimated ± 0.9 pp, high confidence
49	Qwen3.8-27B	84.1%	estimated ± 0.9 pp, high confidence
50	Qwen3.5-122B-A10B	83.9%	measured
51	GPT-5.1	83.4%	estimated ± 0.9 pp, high confidence
52	Claude Opus 4.6 (Adaptive)	83.3%	estimated ± 0.9 pp, high confidence
53	Kimi K2.5	83.3%	estimated ± 0.9 pp, high confidence
54	Kimi K2.5 (Reasoning)	83.3%	estimated ± 0.9 pp, high confidence
55	Step 3.7 Flash	83.2%	estimated ± 0.9 pp, high confidence
56	Gemini 2.5 Pro	82.9%	estimated ± 0.9 pp, high confidence
57	Qwen3.6-27B	82.9%	measured
58	Grok 4.20	82.7%	estimated ± 1.1 pp, medium confidence
59	GPT-5 (medium)	82.4%	estimated ± 0.9 pp, high confidence
60	Muse Glimmer 30B	82.4%	estimated ± 0.9 pp, high confidence
61	GPT-5 (high)	82.3%	estimated ± 0.9 pp, high confidence
62	Qwen3.5-27B	82.3%	measured
63	Claude Opus 4.5 Thinking	82.2%	estimated ± 0.9 pp, high confidence
64	Inkling-Small	82.2%	estimated ± 0.9 pp, high confidence
65	Inkling	81.8%	estimated ± 0.9 pp, high confidence
66	Gemma 4 31B	81.7%	estimated ± 0.9 pp, high confidence
67	Qwen3.6-35B-A3B	81.7%	measured
68	GPT-5.4 mini	81.6%	estimated ± 0.9 pp, high confidence
69	MiMo-V2.6-Flash	81.5%	estimated ± 0.9 pp, high confidence
70	Qwen3.5-35B-A3B	81.4%	measured
71	GLM-5V-Turbo	81.2%	estimated ± 0.9 pp, high confidence
72	Claude Opus 4.6	81.0%	estimated ± 0.9 pp, high confidence
73	GPT-5.1-Codex	81.0%	estimated ± 0.9 pp, high confidence
74	GPT-5.1-Codex-Max	81.0%	estimated ± 0.9 pp, high confidence
75	Claude Opus 4.5	80.0%	estimated ± 0.9 pp, high confidence
76	Claude Sonnet 4.6	79.6%	estimated ± 0.9 pp, high confidence
77	o3	79.2%	estimated ± 0.9 pp, high confidence
78	MiMo-V2-Omni	79.1%	estimated ± 0.9 pp, high confidence
79	Gemma 4 12B	78.9%	estimated ± 0.9 pp, high confidence
80	Gemma 4 26B A4B	78.6%	estimated ± 0.9 pp, high confidence
81	Grok 4	78.3%	estimated ± 0.9 pp, high confidence
82	Claude 4.1 Opus Thinking	77.7%	estimated ± 0.9 pp, high confidence
83	Gemini 2.5 Flash	76.2%	estimated ± 0.9 pp, high confidence
84	GPT-5.4 nano	76.1%	estimated ± 0.9 pp, high confidence
85	Mistral Medium 3.5 128B	75.8%	estimated ± 0.9 pp, high confidence
86	Command A+	75.1%	measured
87	Grok 4.1 Fast (Reasoning)	74.9%	estimated ± 0.9 pp, high confidence
88	Claude 4 Sonnet	74.5%	estimated ± 0.9 pp, high confidence
89	Llama 4 Maverick	74.3%	estimated ± 0.9 pp, high confidence
90	Grok 4 Fast (Reasoning)	74.2%	estimated ± 0.9 pp, high confidence
91	GPT-4.1	73.9%	estimated ± 0.9 pp, high confidence
92	Qwen3-Omni-30B-A3B-Thinking	73.4%	estimated ± 0.9 pp, high confidence
93	GPT-4.1 mini	72.8%	estimated ± 0.9 pp, high confidence
94	Mistral Small 4	72.0%	estimated ± 0.9 pp, high confidence
95	Mistral Small 4 (Reasoning)	72.0%	estimated ± 0.9 pp, high confidence
96	Mistral Large 3	71.7%	estimated ± 0.9 pp, high confidence
97	Qwen3-Omni-30B-A3B-Instruct	71.6%	estimated ± 0.9 pp, high confidence
98	Gemini 1.5 Pro	71.4%	estimated ± 0.9 pp, high confidence
99	Mistral Medium 3	70.8%	estimated ± 0.9 pp, medium confidence
100	Llama 4 Scout	70.8%	estimated ± 0.9 pp, medium confidence
101	Nemotron 3 Nano Omni 30B A3B	70.8%	measured
102	Qwen3.5 397B	70.8%	estimated ± 0.9 pp, medium confidence
103	Qwen3.5 397B (Reasoning)	70.8%	estimated ± 0.9 pp, medium confidence
104	Gemma 4 E4B	70.4%	estimated ± 0.9 pp, medium confidence
105	Grok 4.1 Fast	69.8%	estimated ± 0.9 pp, medium confidence
106	Gemma 3 27B	69.7%	estimated ± 0.9 pp, medium confidence
107	Gemma 4 E2B	69.2%	estimated ± 0.9 pp, medium confidence
108	Nova Pro	69.2%	estimated ± 0.9 pp, medium confidence
109	GPT-4o mini	68.9%	estimated ± 0.9 pp, medium confidence
110	Ternary Bonsai 2 27B	68.8%	estimated ± 6.6 pp, low confidence
111	GPT-4.1 nano	68.8%	estimated ± 0.9 pp, medium confidence
112	Claude 3 Haiku	68.4%	estimated ± 0.9 pp, medium confidence
113	LFM2.5-VL-1.6B-Extract	68.3%	estimated ± 0.9 pp, medium confidence
114	Phi-4 Multimodal Instruct	68.3%	estimated ± 0.9 pp, medium confidence
115	LFM2.5-VL-3B	48.4%	measured
116	ZAYA1-VL-8B	46.0%	measured
117	Interfaze Beta	46.0%	estimated ± 0.4 pp, low confidence
118	North Micro Vision Instruct	37.5%	estimated ± 6.6 pp, low confidence
119	LFM2.5-VL-450M	32.7%	measured

MMMU leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General