// BENCHMARK

MMMU benchmark

AI model leaderboard for the MMMU benchmark. Compare how large language models score on MMMU, see the full ranking, and understand what this AI benchmark measures. Gemini 3.1 Pro currently leads with 81.5%. 11,500 college-level multi-modal questions across 30 disciplines that require visual reasoning.

Leaderboard

# Model Organization Score Variant Source
#1 Gemini 3.1 Pro Google DeepMind 81.5% official ↗
#2 Claude Opus 4.5 Anthropic 80.7% official ↗
#3 o1 OpenAI 77.3% 20241217-val-cited-qvq official ↗
#4 Command A+ Cohere 75.1% official ↗
#5 Muse Spark Meta AI 72.1% official ↗
#6 Claude 3.5 Sonnet Anthropic 70.4% 20241022-val-cited-qvq official ↗
#7 QVQ Alibaba 70.3% val-from-hf-readme official ↗
#8 Magistral Medium 1.1 Mistral AI 70% official ↗
#9 GPT-4o (May 2024) OpenAI 69.1% 20240513-val-cited-qvq official ↗
#10 GPT-4o (Aug 2024) OpenAI 69.1% val-cited-nvlm official ↗
#11 GPT-4o (Aug 2024) OpenAI 68.6% cot-cited-pixtral-large official ↗
#12 Claude 3.5 Sonnet Anthropic 68.4% cot-cited-pixtral-large official ↗
#13 Claude 3.5 Sonnet Anthropic 68.3% val-cited-nvlm official ↗
#14 Gemma 4 31B Google DeepMind 67.2% official ↗
#15 Gemini 1.5 Pro Google DeepMind 66.3% cot-cited-pixtral-large official ↗
#16 Magistral Small 1.1 Mistral AI 66% official ↗
#17 Mistral Medium 3 Mistral AI 65% cited-magistral official ↗
#18 Pixtral Large Mistral AI 64% cot-from-hf-readme official ↗
#19 Mistral Small 3.1 Mistral AI 64% instruct-cited-msmall32 official ↗
#20 Mistral Small 3.2 Mistral AI 62.5% instruct-from-hf-readme official ↗
#21 Gemini 1.5 Pro Google DeepMind 62.2% val-aug-2024-cited-nvlm official ↗
#22 Mistral Small 3.1 Mistral AI 61.9% cited-magistral official ↗
#23 Llama 3.2 90B Meta AI 60.3% val-cited-nvlm official ↗
#24 NVLM-D 72B NVIDIA 59.7% val-megatron-from-hf-readme official ↗
#25 GPT-4o mini OpenAI 59.4% cited-aria official ↗
#26 GPT-4V OpenAI 56.8% val-cited-nvlm official ↗
#27 Aria Rhymes AI 54.9% from-hf-readme official ↗
#28 NVLM-D 72B NVIDIA 54.6% test-megatron-from-hf-readme official ↗
#29 Llama 3.2 90B Meta AI 53.7% cited-pixtral-large official ↗
#30 Pixtral 12B Mistral AI 52.5% cited-aria official ↗
#31 Llama 3.2 11B Meta AI 50.7% cited-aria official ↗

Frequently asked questions about MMMU

What is the MMMU benchmark?

11,500 college-level multi-modal questions across 30 disciplines that require visual reasoning.

How is the MMMU benchmark scored?

MMMU is scored using the accuracy (%) metric, where a higher score is better. The maximum achievable score is 100.000. GenAIList aggregates reported scores from model providers and papers into a single ranked leaderboard.

Which AI model scores highest on MMMU?

As of the latest reported scores on GenAIList, Gemini 3.1 Pro achieves the highest result on MMMU with a score of 81.5%.

Is a higher MMMU score better?

Yes. On MMMU a higher score indicates better performance, so models near the top of the leaderboard are the strongest.