benchgap
Vision & documents

MMMU-Pro leaderboard

As of 2026-11-21, the highest measured score on MMMU-Pro is 85.5% by GPT-6.1 Terra. 108 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#ModelScoreSource
1GPT-6.2 Astra87.8%estimated ± 1.5 pp, medium confidence
2Gemini 4 Argon87.7%estimated ± 5.3 pp, low confidence
3Claude Opus 5.586.6%estimated ± 1.5 pp, medium confidence
4GPT-6 Astra86.0%estimated ± 1.5 pp, medium confidence
5Claude Fable 5.185.8%estimated ± 6.9 pp, low confidence
6GPT-6.2 Sol85.8%estimated ± 5.3 pp, low confidence
7Gemini 4 Neon85.5%estimated ± 5.3 pp, low confidence
8GPT-6.1 Terra85.5%measured
9GPT-6.1 Sol85.3%estimated ± 1.5 pp, medium confidence
10Gemini 3.8 Flash85.0%estimated ± 1.5 pp, medium confidence
11Gemini 3.7 Flash85.0%estimated ± 1.5 pp, medium confidence
12Claude Opus 584.4%estimated ± 1.5 pp, high confidence
13Kimi K3.1 Pro84.1%measured
14Gemini 3.1 Pro83.9%measured
15Gemini 3.5 Flash83.6%measured
16Gemini 3.6 Flash83.2%estimated ± 1.5 pp, high confidence
17GPT-5.6 Sol83.0%measured
18GPT-6 Sol83.0%estimated ± 1.5 pp, high confidence
19Qwen3.8 Max Preview82.9%estimated ± 1.5 pp, high confidence
20Claude Mythos 582.3%estimated ± 3.1 pp, high confidence
21Qwen3.8 Max82.3%measured
22Glimpse82.2%estimated ± 3.1 pp, high confidence
23Muse Spark 1.181.9%estimated ± 2.5 pp, medium confidence
24Muse Ember81.8%estimated ± 5.3 pp, low confidence
25Claude Fable 581.6%estimated ± 6.9 pp, medium confidence
26Inkling 2 Small81.6%measured
27Kimi K381.6%measured
28Seed 2.1 Pro81.6%measured
29Claude Sonnet 5.581.4%estimated ± 5.3 pp, low confidence
30Muse Spark Open81.3%estimated ± 1.5 pp, high confidence
31LongCat-2.581.3%estimated ± 3.1 pp, high confidence
32Claude Opus 4.881.2%estimated ± 3.1 pp, high confidence
33GPT-5.481.2%measured
34GPT-5.581.2%measured
35Gemini 3 Pro81.0%measured
36Grok 4.580.9%estimated ± 1.5 pp, high confidence
37GPT-5.6 Terra80.7%measured
38Muse Spark80.4%measured
39GPT-6 Luna80.3%estimated ± 1.5 pp, high confidence
40Sakana Fugu-Ultra80.2%estimated ± 3.1 pp, high confidence
41Seed 2.1 Turbo80.1%measured
42Qwen3.8-Omni-Flash80.1%estimated ± 3.0 pp, high confidence
43Apodex 1.179.9%estimated ± 1.5 pp, high confidence
44Apodex 1.1 Mini79.9%estimated ± 1.5 pp, high confidence
45MiniMax M3.1 Flash79.8%estimated ± 1.5 pp, high confidence
46Gemini 3.5 Flash-Lite79.7%estimated ± 1.5 pp, high confidence
47Ling 3.0 Flash VL79.7%estimated ± 1.5 pp, high confidence
48Sakana Fugu79.7%estimated ± 3.1 pp, high confidence
49Claude Opus 4.779.6%estimated ± 1.5 pp, high confidence
50GPT-5.279.5%measured
51Kimi K2.679.4%measured
52Gemini 3 Flash79.4%estimated ± 1.5 pp, high confidence
53Qwen3.8-Flash-Next79.4%estimated ± 1.5 pp, high confidence
54GPT-5.3 Codex79.3%estimated ± 1.5 pp, high confidence
55dots3-note Preview79.1%measured
56Qwen3.5 397B79.0%measured
57Qwen3.7 Plus79.0%measured
58Claude Haiku 5.578.9%estimated ± 5.3 pp, low confidence
59Qwen3.6 Plus78.8%measured
60Hy478.6%measured
61Kimi K2.578.5%measured
62GPT-5.6 Luna78.4%measured
63Claude Sonnet 578.3%estimated ± 1.5 pp, high confidence
64Grok 4.378.1%measured
65MiniMax M378.1%measured
66Pareto 26.978.0%measured
67DeepSeek V4.1 Flash 101778.0%estimated ± 1.5 pp, high confidence
68MiMo-V2.577.9%measured
69GLM-5.3-Flash77.9%estimated ± 2.5 pp, medium confidence
70GPT-5.2-Codex77.4%estimated ± 1.5 pp, high confidence
71Qwen3.8-27B77.4%estimated ± 1.5 pp, high confidence
72Claude Opus 4.677.3%measured
73Gemma 4 31B76.9%measured
74Qwen3.5-122B-A10B76.9%measured
75GPT-5.176.6%estimated ± 1.5 pp, high confidence
76GPT-5.4 mini76.6%measured
77Claude Opus 4.6 (Adaptive)76.6%estimated ± 1.5 pp, high confidence
78Step 3.7 Flash76.5%estimated ± 1.5 pp, high confidence
79Gemini 2.5 Pro76.1%estimated ± 1.5 pp, high confidence
80Step 5 Preview76.0%measured
81Hy4 flash75.8%measured
82Qwen3.6-27B75.8%measured
83GPT-575.5%estimated ± 1.5 pp, high confidence
84Qwen3.6-35B-A3B75.3%measured
85Claude Opus 4.5 Thinking75.3%estimated ± 1.5 pp, high confidence
86Gemini 3.1 Flash-Lite75.2%estimated ± 3.1 pp, high confidence
87Grok 4.2075.2%measured
88Qwen3.5-35B-A3B75.1%measured
89Qwen3.5-27B75.0%measured
90Inkling 2 Mini74.9%measured
91Holo2-235B-A22B74.5%estimated ± 8.7 pp, low confidence
92MiMo-V2.6-Flash74.4%estimated ± 1.5 pp, high confidence
93GLM-5V-Turbo74.1%estimated ± 1.5 pp, high confidence
94Inkling-Small74.0%measured
95Muse Glimmer 30B74.0%measured
96GPT-5.1-Codex73.9%estimated ± 1.5 pp, high confidence
97GPT-5.1-Codex-Max73.9%estimated ± 1.5 pp, high confidence
98Gemma 4 26B A4B73.8%measured
99Nemotron 3.5 Super VL 120B A12B73.6%estimated ± 2.5 pp, low confidence
100Nemotron 3.5 Super VL 120B A12B NVFP473.6%estimated ± 2.5 pp, low confidence
101Inkling73.5%measured
102Holo2-30B-A3B72.2%estimated ± 8.7 pp, low confidence
103Claude Sonnet 4.672.0%estimated ± 1.5 pp, high confidence
104Claude Haiku 4.5 Thinking71.9%estimated ± 6.9 pp, medium confidence
105o371.5%estimated ± 1.5 pp, high confidence
106MiMo-V2-Omni71.3%estimated ± 1.5 pp, high confidence
107Interfaze Beta71.1%measured
108Claude Haiku 4.571.0%estimated ± 6.9 pp, medium confidence
109o4-mini70.6%estimated ± 1.5 pp, high confidence
110Claude Opus 4.570.6%measured
111Grok 470.3%estimated ± 1.5 pp, high confidence
112Gemma 4 12B69.1%measured
113Holo2-8B68.3%estimated ± 8.7 pp, low confidence
114Qwen3-VL-235B-A22B Instruct68.1%measured
115Holo2-4B67.3%estimated ± 8.7 pp, low confidence
116Gemini 2.5 Flash66.9%estimated ± 1.5 pp, high confidence
117Qwen3.5-4B66.8%estimated ± 1.5 pp, high confidence
118Amazon Nova 2 Lite66.3%estimated ± 6.9 pp, medium confidence
119Mistral Medium 3.5 128B66.3%estimated ± 1.5 pp, high confidence
120GPT-5.4 nano66.1%measured
121GLM-4.6V66.0%measured
122Claude 4 Opus Thinking65.8%estimated ± 2.2 pp, high confidence
123Grok 4.1 Fast (Reasoning)64.6%estimated ± 1.5 pp, high confidence
124Claude 3.7 Sonnet Thinking63.8%estimated ± 2.2 pp, high confidence
125Claude 4 Sonnet63.7%estimated ± 1.5 pp, high confidence
126Llama 4 Maverick63.3%estimated ± 1.5 pp, high confidence
127Grok 4 Fast (Reasoning)63.0%estimated ± 1.5 pp, high confidence
128Command A+63.0%measured
129GPT-4.162.4%estimated ± 1.5 pp, high confidence
130Claude 4 Opus62.0%estimated ± 2.2 pp, high confidence
131Qwen3-Omni-30B-A3B-Thinking61.3%estimated ± 1.5 pp, high confidence
132Claude 3.7 Sonnet61.2%estimated ± 1.5 pp, high confidence
133GLM-4.6V-Flash60.6%measured
134GPT-4.1 mini59.7%estimated ± 1.5 pp, high confidence
135Mistral Small 457.7%estimated ± 1.5 pp, high confidence
136Mistral Large 356.5%estimated ± 1.5 pp, high confidence
137Gemini 2.0 Flash56.3%estimated ± 2.2 pp, high confidence
138Qwen3-Omni-30B-A3B-Instruct56.2%estimated ± 1.5 pp, high confidence
139Gemini 1.5 Pro55.7%estimated ± 1.5 pp, high confidence
140Gemini 2.5 Flash-Lite54.6%estimated ± 1.5 pp, high confidence
141Nemotron 3 Nano Omni 30B A3B53.7%estimated ± 1.5 pp, high confidence
142Mistral Medium 353.5%estimated ± 1.5 pp, high confidence
143Llama 4 Scout53.4%estimated ± 1.5 pp, high confidence
144MiniMax-VL-0152.7%measured
145Gemma 4 E4B52.6%measured
146Grok 252.4%estimated ± 2.2 pp, high confidence
147Gemini 2.0 Flash-Lite51.2%estimated ± 2.2 pp, high confidence
148Qwen2.5-VL-32B49.5%measured
149Mistral Small 3.149.3%measured
150Grok 2 mini49.0%estimated ± 2.2 pp, high confidence
151Gemini 1.5 Flash48.6%estimated ± 1.5 pp, high confidence
152Grok 4.1 Fast48.6%estimated ± 1.5 pp, high confidence
153Gemma 3 27B48.1%estimated ± 1.5 pp, high confidence
154LFM3-VL-2B45.7%measured
155Claude 3.5 Haiku45.6%estimated ± 1.5 pp, high confidence
156Ternary Bonsai 2 27B44.6%estimated ± 7.6 pp, medium confidence
157Gemma 4 E2B44.2%measured
158Nova Pro44.2%estimated ± 1.5 pp, medium confidence
159GPT-4o mini41.3%estimated ± 1.5 pp, medium confidence
160GPT-4.1 nano39.9%estimated ± 1.5 pp, medium confidence
161ZAYA1-VL-8B31.6%estimated ± 2.2 pp, medium confidence
162Claude 3 Haiku31.0%estimated ± 1.5 pp, medium confidence
163LFM2.5-VL-3B30.5%measured
164LFM2.5-VL-1.6B-Extract27.4%estimated ± 1.5 pp, medium confidence
165North Micro Vision Instruct25.8%estimated ± 10.8 pp, low confidence
166LFM2.5-VL-450M20.5%estimated ± 2.2 pp, medium confidence
167Phi-4 Multimodal Instruct18.8%estimated ± 1.5 pp, medium confidence

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General