benchgap
Vision & documents

MMMU-Pro w/ Python leaderboard

As of 2026-11-21, the highest measured score on MMMU-Pro w/ Python is 90.0% by Kimi K3.1 Pro. 119 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Astra91.3%estimated ± 2.0 pp, medium confidence
2Kimi K3.1 Pro90.0%measured
3Claude Opus 5.589.7%estimated ± 2.0 pp, medium confidence
4GPT-6 Astra88.9%estimated ± 2.0 pp, medium confidence
5GPT-6.1 Sol88.1%estimated ± 2.0 pp, medium confidence
6Gemini 3.8 Flash87.8%estimated ± 2.0 pp, medium confidence
7Gemini 3.7 Flash87.7%estimated ± 2.0 pp, medium confidence
8Claude Opus 587.0%estimated ± 2.0 pp, medium confidence
9GPT-6.1 Terra87.0%estimated ± 2.0 pp, medium confidence
10Gemini 3.5 Flash86.6%estimated ± 2.0 pp, medium confidence
11Claude Fable 5.185.7%estimated ± 2.7 pp, low confidence
12Gemini 3.6 Flash85.6%estimated ± 2.0 pp, high confidence
13GPT-6 Sol85.3%estimated ± 2.0 pp, high confidence
14Qwen3.8 Max Preview85.2%estimated ± 2.0 pp, high confidence
15Gemini 3.1 Pro84.9%estimated ± 2.0 pp, high confidence
16GPT-5.6 Sol84.6%measured
17Inkling 2 Small84.2%estimated ± 2.0 pp, high confidence
18Muse Spark Open83.5%estimated ± 2.0 pp, high confidence
19Kimi K383.4%measured
20GPT-5.583.2%measured
21Qwen3.8 Max83.2%estimated ± 2.6 pp, high confidence
22Muse Spark83.2%estimated ± 2.0 pp, high confidence
23Qwen3.7 Plus83.2%estimated ± 2.0 pp, high confidence
24Claude Fable 583.1%estimated ± 2.7 pp, medium confidence
25Grok 4.583.1%estimated ± 2.0 pp, high confidence
26GLM-5.3-Flash83.0%estimated ± 2.7 pp, medium confidence
27Claude Sonnet 5.582.9%estimated ± 2.7 pp, medium confidence
28Gemini 3 Pro82.9%estimated ± 2.0 pp, high confidence
29Seed 2.1 Pro82.8%estimated ± 2.6 pp, high confidence
30GPT-6 Luna82.5%estimated ± 2.0 pp, high confidence
31GPT-5.482.1%measured
32Apodex 1.182.0%estimated ± 2.0 pp, high confidence
33Apodex 1.1 Mini82.0%estimated ± 2.0 pp, high confidence
34GPT-5.6 Terra82.0%measured
35MiniMax M3.1 Flash81.9%estimated ± 2.0 pp, high confidence
36Seed 2.1 Turbo81.9%estimated ± 2.6 pp, high confidence
37Gemini 3.5 Flash-Lite81.8%estimated ± 2.0 pp, high confidence
38Ling 3.0 Flash VL81.8%estimated ± 2.0 pp, high confidence
39Claude Opus 4.781.6%estimated ± 2.0 pp, high confidence
40GPT-5.281.5%estimated ± 2.6 pp, high confidence
41Gemini 3 Flash81.5%estimated ± 2.0 pp, high confidence
42MiniMax M381.5%estimated ± 2.0 pp, high confidence
43Qwen3.8-Flash-Next81.5%estimated ± 2.0 pp, high confidence
44GPT-5.3 Codex81.4%estimated ± 2.0 pp, high confidence
45dots3-note Preview81.3%estimated ± 2.6 pp, high confidence
46Grok 4.381.0%estimated ± 2.0 pp, high confidence
47Hy481.0%estimated ± 2.6 pp, high confidence
48Qwen3.6 Plus80.9%estimated ± 2.0 pp, high confidence
49Pareto 26.980.6%estimated ± 2.6 pp, high confidence
50MiMo-V2.580.5%estimated ± 2.6 pp, high confidence
51Claude Opus 4.880.3%estimated ± 2.7 pp, medium confidence
52Claude Sonnet 580.3%estimated ± 2.0 pp, high confidence
53Qwen3.5 397B80.3%estimated ± 2.0 pp, high confidence
54Kimi K2.680.1%measured
55DeepSeek V4.1 Flash 101780.0%estimated ± 2.0 pp, high confidence
56GPT-5.6 Luna79.5%measured
57Step 5 Preview79.5%estimated ± 2.0 pp, high confidence
58GPT-5.2-Codex79.4%estimated ± 2.0 pp, high confidence
59Qwen3.8-27B79.4%estimated ± 2.0 pp, high confidence
60Hy4 flash79.2%estimated ± 2.6 pp, high confidence
61Grok 4.2078.8%estimated ± 2.6 pp, high confidence
62GPT-5.178.7%estimated ± 2.0 pp, high confidence
63Inkling 2 Mini78.6%estimated ± 2.6 pp, high confidence
64Claude Opus 4.6 (Adaptive)78.6%estimated ± 2.0 pp, high confidence
65Kimi K2.578.6%estimated ± 2.0 pp, high confidence
66Step 3.7 Flash78.5%estimated ± 2.0 pp, high confidence
67Qwen3.5-122B-A10B78.2%estimated ± 2.0 pp, high confidence
68Qwen3.5-27B78.2%estimated ± 2.0 pp, high confidence
69Qwen3.6-35B-A3B78.2%estimated ± 2.0 pp, high confidence
70Gemini 2.5 Pro78.1%estimated ± 2.0 pp, high confidence
71GPT-5.4 mini78.0%measured
72Gemini 3.1 Flash-Lite77.9%estimated ± 2.7 pp, medium confidence
73Qwen3.6-27B77.9%estimated ± 2.0 pp, high confidence
74Claude Haiku 5.577.6%estimated ± 2.7 pp, medium confidence
75Muse Glimmer 30B77.6%estimated ± 2.0 pp, high confidence
76GPT-577.5%estimated ± 2.0 pp, high confidence
77Claude Opus 4.5 Thinking77.3%estimated ± 2.0 pp, high confidence
78Inkling-Small77.3%estimated ± 2.0 pp, high confidence
79Inkling76.9%estimated ± 2.0 pp, high confidence
80Gemma 4 31B76.8%estimated ± 2.0 pp, high confidence
81MiMo-V2.6-Flash76.5%estimated ± 2.0 pp, high confidence
82GLM-5V-Turbo76.2%estimated ± 2.0 pp, high confidence
83Interfaze Beta76.2%estimated ± 2.6 pp, high confidence
84Qwen3.5-35B-A3B76.2%estimated ± 2.0 pp, high confidence
85Claude Opus 4.676.0%estimated ± 2.0 pp, high confidence
86GPT-5.1-Codex76.0%estimated ± 2.0 pp, high confidence
87GPT-5.1-Codex-Max76.0%estimated ± 2.0 pp, high confidence
88Claude Haiku 4.5 Thinking75.2%estimated ± 2.7 pp, medium confidence
89Claude Opus 4.574.8%estimated ± 2.0 pp, high confidence
90Claude Sonnet 4.674.3%estimated ± 2.0 pp, high confidence
91Claude Haiku 4.574.2%estimated ± 2.7 pp, medium confidence
92Qwen3-VL-235B-A22B Instruct74.1%estimated ± 2.6 pp, high confidence
93o373.8%estimated ± 2.0 pp, high confidence
94MiMo-V2-Omni73.6%estimated ± 2.0 pp, high confidence
95Gemma 4 12B73.5%estimated ± 2.0 pp, high confidence
96Gemma 4 26B A4B73.0%estimated ± 2.0 pp, high confidence
97o4-mini73.0%estimated ± 2.0 pp, high confidence
98GLM-4.6V72.7%estimated ± 2.6 pp, medium confidence
99Grok 472.6%estimated ± 2.0 pp, high confidence
100Gemini 2.5 Flash69.7%estimated ± 2.0 pp, high confidence
101Qwen3.5-4B69.6%estimated ± 2.0 pp, high confidence
102GPT-5.4 nano69.5%measured
103Mistral Medium 3.5 128B69.1%estimated ± 2.0 pp, medium confidence
104GLM-4.6V-Flash68.8%estimated ± 2.6 pp, medium confidence
105Amazon Nova 2 Lite68.7%estimated ± 2.7 pp, low confidence
106Grok 4.1 Fast (Reasoning)67.7%estimated ± 2.0 pp, medium confidence
107Command A+67.6%estimated ± 2.0 pp, medium confidence
108Claude 4 Sonnet66.9%estimated ± 2.0 pp, medium confidence
109Llama 4 Maverick66.6%estimated ± 2.0 pp, medium confidence
110Grok 4 Fast (Reasoning)66.4%estimated ± 2.0 pp, medium confidence
111GPT-4.165.8%estimated ± 2.0 pp, medium confidence
112Qwen3-Omni-30B-A3B-Thinking64.9%estimated ± 2.0 pp, medium confidence
113Claude 3.7 Sonnet64.8%estimated ± 2.0 pp, medium confidence
114GPT-4.1 mini63.6%estimated ± 2.0 pp, medium confidence
115MiniMax-VL-0162.6%estimated ± 2.6 pp, medium confidence
116Mistral Small 461.9%estimated ± 2.0 pp, medium confidence
117Mistral Large 360.9%estimated ± 2.0 pp, medium confidence
118Qwen3-Omni-30B-A3B-Instruct60.7%estimated ± 2.0 pp, medium confidence
119Gemini 1.5 Pro60.2%estimated ± 2.0 pp, medium confidence
120Qwen2.5-VL-32B60.0%estimated ± 2.6 pp, medium confidence
121Mistral Small 3.159.8%estimated ± 2.6 pp, medium confidence
122Gemini 2.5 Flash-Lite59.3%estimated ± 2.0 pp, medium confidence
123Nemotron 3 Nano Omni 30B A3B58.6%estimated ± 2.0 pp, medium confidence
124Mistral Medium 358.4%estimated ± 2.0 pp, medium confidence
125Llama 4 Scout58.3%estimated ± 2.0 pp, medium confidence
126Gemma 4 E4B57.0%estimated ± 2.0 pp, medium confidence
127Gemini 1.5 Flash54.3%estimated ± 2.0 pp, medium confidence
128Grok 4.1 Fast54.3%estimated ± 2.0 pp, medium confidence
129Gemma 3 27B53.9%estimated ± 2.0 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General