MMMU
MultimodalMassive Multi-discipline Multimodal Understanding — 11,500 questions from college exams across 30 subjects and 6 disciplines, requiring college-level reasoning over images.
Models Tested
3
Best Score
81.7%
Median Score
73.4%
Scoring: accuracy
Introduced: 2023-11
Leaderboard3 models
| # | Model | Developer | Score |
|---|---|---|---|
| 🥇 | Gemini 2.5 Pro | Google DeepMind | 81.7% |
| 🥈 | Llama 4 Maverick | Meta AI (FAIR) | 73.4% |
| 🥉 | Llama 4 Scout | Meta AI (FAIR) | 69.4% |