benchgap
Vision & documents

MMMU leaderboard

As of 2026-11-21, the highest measured score on MMMU is 87.2% by Gemini 3 Pro. 119 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1Claude Mythos 593.1%estimated ± 3.7 pp, low confidence
2Glimpse92.7%estimated ± 3.7 pp, low confidence
3GPT-6.2 Astra91.9%estimated ± 2.7 pp, medium confidence
4Claude Opus 5.590.9%estimated ± 2.7 pp, medium confidence
5LongCat-2.590.7%estimated ± 3.7 pp, low confidence
6Claude Opus 4.890.5%estimated ± 3.7 pp, low confidence
7GPT-6 Astra90.4%estimated ± 2.7 pp, medium confidence
8GPT-6.1 Sol89.9%estimated ± 2.7 pp, medium confidence
9Gemini 3.8 Flash89.7%estimated ± 2.7 pp, medium confidence
10Gemini 3.7 Flash89.7%estimated ± 2.7 pp, medium confidence
11Muse Spark 1.189.4%estimated ± 3.7 pp, low confidence
12Claude Opus 589.2%estimated ± 2.7 pp, medium confidence
13GPT-6.1 Terra89.2%estimated ± 0.8 pp, medium confidence
14Gemini 3.5 Flash88.6%estimated ± 0.8 pp, medium confidence
15Kimi K3.1 Pro88.5%estimated ± 0.8 pp, medium confidence
16Gemini 3.6 Flash88.3%estimated ± 2.7 pp, medium confidence
17Qwen3.8 Max88.2%estimated ± 1.7 pp, medium confidence
18GPT-6 Sol88.2%estimated ± 2.7 pp, medium confidence
19Qwen3.8 Max Preview88.1%estimated ± 2.7 pp, medium confidence
20GPT-5.6 Sol88.1%estimated ± 0.8 pp, medium confidence
21Sakana Fugu-Ultra88.1%estimated ± 3.7 pp, low confidence
22Gemini 3.1 Pro88.0%estimated ± 0.8 pp, medium confidence
23Seed 2.1 Pro87.7%estimated ± 1.7 pp, medium confidence
24Gemini 3 Pro87.2%measured
25Inkling 2 Small87.1%estimated ± 0.8 pp, medium confidence
26Sakana Fugu87.1%estimated ± 3.7 pp, low confidence
27Muse Spark Open87.0%estimated ± 2.7 pp, medium confidence
28Kimi K386.7%estimated ± 0.8 pp, medium confidence
29Grok 4.586.7%estimated ± 2.7 pp, medium confidence
30GPT-5.286.7%measured
31GPT-5.6 Terra86.6%estimated ± 0.8 pp, medium confidence
32Seed 2.1 Turbo86.5%estimated ± 1.7 pp, high confidence
33Muse Spark86.4%estimated ± 0.8 pp, medium confidence
34GPT-5.586.4%estimated ± 0.8 pp, medium confidence
35GPT-6 Luna86.3%estimated ± 2.7 pp, high confidence
36Qwen3.7 Plus86.1%estimated ± 0.8 pp, medium confidence
37Qwen3.6 Plus86.0%measured
38Apodex 1.186.0%estimated ± 2.7 pp, high confidence
39Apodex 1.1 Mini86.0%estimated ± 2.7 pp, high confidence
40MiniMax M3.1 Flash85.9%estimated ± 2.7 pp, high confidence
41GPT-5.485.9%estimated ± 0.8 pp, medium confidence
42Gemini 3.5 Flash-Lite85.9%estimated ± 2.7 pp, high confidence
43Ling 3.0 Flash VL85.9%estimated ± 2.7 pp, high confidence
44Kimi K2.685.8%estimated ± 0.8 pp, high confidence
45dots3-note Preview85.8%estimated ± 1.7 pp, high confidence
46Qwen3.8-Omni-Flash85.8%estimated ± 3.7 pp, medium confidence
47Claude Opus 4.785.7%estimated ± 2.7 pp, high confidence
48Gemini 3 Flash85.6%estimated ± 2.7 pp, high confidence
49Qwen3.8-Flash-Next85.6%estimated ± 2.7 pp, high confidence
50GPT-5.3 Codex85.6%estimated ± 2.7 pp, high confidence
51Hy485.4%estimated ± 1.7 pp, high confidence
52GPT-5.6 Luna85.2%estimated ± 0.8 pp, high confidence
53MiniMax M385.2%estimated ± 0.8 pp, high confidence
54GLM-5.3-Flash85.1%estimated ± 2.9 pp, low confidence
55Qwen3.5 397B85.0%measured
56Grok 4.385.0%estimated ± 0.8 pp, high confidence
57Pareto 26.984.9%estimated ± 1.7 pp, high confidence
58MiMo-V2.584.9%estimated ± 1.7 pp, high confidence
59Claude Sonnet 584.8%estimated ± 2.7 pp, high confidence
60DeepSeek V4.1 Flash 101784.6%estimated ± 2.7 pp, high confidence
61Kimi K2.584.3%measured
62GPT-5.2-Codex84.2%estimated ± 2.7 pp, high confidence
63Qwen3.8-27B84.2%estimated ± 2.7 pp, high confidence
64Qwen3.5-122B-A10B83.9%measured
65Step 5 Preview83.8%estimated ± 0.8 pp, high confidence
66GPT-5.183.7%estimated ± 2.7 pp, high confidence
67Claude Opus 4.6 (Adaptive)83.6%estimated ± 2.7 pp, high confidence
68Step 3.7 Flash83.6%estimated ± 2.7 pp, high confidence
69Gemini 2.5 Pro83.3%estimated ± 2.7 pp, high confidence
70Hy4 flash83.2%estimated ± 1.7 pp, high confidence
71Gemma 4 31B82.9%estimated ± 0.8 pp, high confidence
72Qwen3.6-27B82.9%measured
73GPT-582.9%estimated ± 2.7 pp, high confidence
74GPT-5.4 mini82.8%estimated ± 0.8 pp, high confidence
75Grok 4.2082.8%estimated ± 1.7 pp, high confidence
76Claude Opus 4.5 Thinking82.8%estimated ± 2.7 pp, high confidence
77Claude Opus 4.682.7%estimated ± 0.8 pp, high confidence
78Inkling 2 Mini82.6%estimated ± 1.7 pp, high confidence
79Muse Glimmer 30B82.5%estimated ± 0.8 pp, high confidence
80Nemotron 3.5 Super VL 120B A12B82.4%estimated ± 3.7 pp, high confidence
81Inkling-Small82.4%estimated ± 0.8 pp, high confidence
82Qwen3.5-27B82.3%measured
83Nemotron 3.5 Super VL 120B A12B NVFP482.2%estimated ± 3.7 pp, high confidence
84MiMo-V2.6-Flash82.2%estimated ± 2.7 pp, high confidence
85Inkling82.0%estimated ± 0.8 pp, high confidence
86GLM-5V-Turbo82.0%estimated ± 2.7 pp, high confidence
87GPT-5.1-Codex81.8%estimated ± 2.7 pp, high confidence
88GPT-5.1-Codex-Max81.8%estimated ± 2.7 pp, high confidence
89Qwen3.6-35B-A3B81.7%measured
90Qwen3.5-35B-A3B81.4%measured
91Claude Opus 4.580.7%measured
92Claude Sonnet 4.680.5%estimated ± 2.7 pp, high confidence
93Gemma 4 26B A4B80.4%estimated ± 0.8 pp, high confidence
94o380.2%estimated ± 2.7 pp, high confidence
95MiMo-V2-Omni80.1%estimated ± 2.7 pp, high confidence
96Gemini 3.1 Flash-Lite80.0%estimated ± 3.7 pp, medium confidence
97Interfaze Beta79.6%estimated ± 1.7 pp, high confidence
98o4-mini79.6%estimated ± 2.7 pp, high confidence
99Grok 479.3%estimated ± 2.7 pp, high confidence
100Gemma 4 12B79.3%estimated ± 0.8 pp, high confidence
101Qwen3-VL-235B-A22B Instruct78.7%measured
102Gemini 2.5 Flash77.0%estimated ± 2.7 pp, high confidence
103Qwen3.5-4B77.0%estimated ± 2.7 pp, high confidence
104GPT-5.4 nano76.6%estimated ± 0.8 pp, high confidence
105Mistral Medium 3.5 128B76.6%estimated ± 2.7 pp, high confidence
106Claude 4 Opus Thinking76.5%measured
107GLM-4.6V76.0%measured
108Ternary Bonsai 2 27B75.8%estimated ± 9.0 pp, medium confidence
109Grok 4.1 Fast (Reasoning)75.5%estimated ± 2.7 pp, high confidence
110Command A+75.1%measured
111Claude 3.7 Sonnet Thinking75.0%measured
112Claude 4 Sonnet74.8%estimated ± 2.7 pp, high confidence
113Llama 4 Maverick74.6%estimated ± 2.7 pp, high confidence
114Grok 4 Fast (Reasoning)74.4%estimated ± 2.7 pp, high confidence
115GPT-4.173.9%estimated ± 2.7 pp, high confidence
116Claude 4 Opus73.7%measured
117Qwen3-Omni-30B-A3B-Thinking73.2%estimated ± 2.7 pp, high confidence
118Gemini 2.5 Flash-Lite72.9%measured
119GPT-4.1 mini72.0%estimated ± 2.7 pp, high confidence
120Claude 3.7 Sonnet71.8%measured
121GLM-4.6V-Flash71.1%measured
122Nemotron 3 Nano Omni 30B A3B70.8%measured
123Mistral Small 470.5%estimated ± 2.7 pp, high confidence
124Mistral Large 369.7%estimated ± 2.7 pp, high confidence
125Qwen3-Omni-30B-A3B-Instruct69.5%estimated ± 2.7 pp, high confidence
126Gemini 2.0 Flash69.3%measured
127Gemini 1.5 Pro69.1%estimated ± 2.7 pp, high confidence
128MiniMax-VL-0168.5%measured
129Mistral Medium 367.5%estimated ± 2.7 pp, high confidence
130Llama 4 Scout67.4%estimated ± 2.7 pp, high confidence
131Grok 266.1%measured
132Gemma 4 E4B65.9%estimated ± 0.8 pp, medium confidence
133Gemini 2.0 Flash-Lite65.1%measured
134Mistral Small 3.164.0%measured
135Grok 4.1 Fast63.5%estimated ± 2.7 pp, high confidence
136Grok 2 mini63.2%measured
137Gemma 3 27B63.2%estimated ± 2.7 pp, medium confidence
138Qwen2.5-VL-32B63.0%estimated ± 1.7 pp, high confidence
139Claude 3.5 Haiku61.0%estimated ± 2.7 pp, medium confidence
140Nova Pro59.8%estimated ± 2.7 pp, medium confidence
141Gemma 4 E2B59.0%estimated ± 1.7 pp, high confidence
142Gemini 1.5 Flash58.3%measured
143GPT-4o mini57.1%estimated ± 2.7 pp, medium confidence
144GPT-4.1 nano55.8%estimated ± 2.7 pp, medium confidence
145LFM3-VL-2B55.7%measured
146LFM2.5-VL-3B48.4%measured
147ZAYA1-VL-8B46.0%measured
148Claude 3 Haiku45.9%estimated ± 2.7 pp, medium confidence
149LFM2.5-VL-1.6B-Extract40.8%estimated ± 2.7 pp, medium confidence
150North Micro Vision Instruct38.4%estimated ± 9.0 pp, medium confidence
151LFM2.5-VL-450M32.7%measured

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General