MMMU
College-level multi-discipline questions over images
multimodalactive
Definition
- Version
- not in the notes
- Category
- multimodal
- Maintainer
- MMMU team (OSU/Waterloo et al.), USA/Canada
- Status
- active
- Home page
- https://arxiv.org/abs/2311.16502
Sources
- huggingface.co/microsoft/Phi-4-reasoning-vision-15B https://huggingface.co/microsoft/Phi-4-reasoning-vision-15B
- MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI arxiv.org https://arxiv.org/abs/2311.16502
- MMMU Leaderboard & Scores , October 2026 benchlm.ai https://www.benchlm.ai/benchmarks/mmmu
Scores
| Model | Score | Who ran it | Setting | View or population | Date | Source |
|---|---|---|---|---|---|---|
 Claude Fable 5.1 | 90.64 % | independent | Vals mirror | Vals.ai mirror shown on BenchLM | 10 Oct 2026 | benchlm.ai |
 Gemini 3 Pro | 87.2 % | vendor-reported | – | rank 1, BenchLM aggregator, top score | 10 Oct 2026 | benchlm.ai |
| Claude Opus 4 | 73.7 % | vendor-reported | Without extended thinking | Anthropic announcement | 22 May 2025 | anthropic.com |
 Pixtral Large | 64 % | vendor-reported | CoT | Pixtral Large model card | 18 Nov 2024 | huggingface.co |
 BAGEL (BAGEL-7B-MoT) | 55.3 % | vendor-reported | – | BAGEL model card | May 2025 | huggingface.co |
 Phi-4-reasoning-vision-15B | 54.3 % | vendor-reported | – | – | – | huggingface.co |
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "MMMU", https://aurion.semai.id/benchmarks/mmmu/, records of October 2026.