multimodal benchmark
MMMU-Pro
A more robust multi-discipline multimodal understanding benchmark that enhances MMMU through a three-step process: filtering text-only answerable questions, augmenting candidate options, and introducing vision-only input settings. Achieves significantly lower model performance (16.8-26.9%) compared to original MMMU, providing more rigorous evaluation that closely mimics real-world scenarios.
Updated Aug 11, 2026
Models66
Model coverage66
MetricScore
EvidenceB
MMMU-Pro Ranking
Higher score ranks better on this benchmark.
MMMU-Pro Score Distribution
A closer view of the leading scores on this benchmark.
MMMU-Pro Highlights
The leading models and scores on this benchmark.
What is MMMU-Pro?
What MMMU-Pro measures and how its scores work.
A more robust multi-discipline multimodal understanding benchmark that enhances MMMU through a three-step process: filtering text-only answerable questions, augmenting candidate options, and introducing vision-only input settings. Achieves significantly lower model performance (16.8-26.9%) compared to original MMMU, providing more rigorous evaluation that closely mimics real-world scenarios.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- MMMU-Pro
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- mmmu-pro|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about MMMU-Pro.
Which model scores highest on MMMU-Pro?
Gemini 3.5 Flash is currently ranked first with 83.6%.
What does MMMU-Pro measure?
A more robust multi-discipline multimodal understanding benchmark that enhances MMMU through a three-step process: filtering text-only answerable questions, augmenting candidate options, and introducing vision-only input settings. Achieves significantly lower model performance (16.8-26.9%) compared to original MMMU, providing more rigorous evaluation that closely mimics real-world scenarios.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
66 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.