MMMU Pro: leaderboard
1,730 college-level image questions across 30 subjects in 6 disciplines (MMMU team, 2024), filtered to need the image, with 10 options and a vision-only setting that embeds the question in the image.
Metric: Accuracy (%). Source: www.vals.ai. Status: saturation imminent. 93 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Fable 5.1 (Max) | 90.64 |
| 2 | Claude Opus 5 (Max) | 89.88 |
| 3 | Claude Fable 5 (Max) | 89.31 |
| 4 | Gemini 3.8 Flash (High) | 89.08 |
| 5 | Gemini 3.7 Flash (High) | 88.96 |
| 6 | GPT-5.6 Sol (Max) | 88.84 |
| 7 | Gemini 3.6 Flash (High) | 88.38 |
| 8 | GPT-5.5 (xHigh) | 88.27 |
| 9 | Gemini 3.5 Flash (High) | 88.27 |
| 10 | Gemini 3.1 Pro (Preview) (High) | 88.21 |
| 11 | Kimi K3 | 88.15 |
| 12 | Qwen 3.8 Max | 88.03 |
| 13 | Gemini 3 Flash (Preview) (High) | 87.63 |
| 14 | GPT-5.4 (xHigh) | 87.51 |
| 15 | Gemini 3 Pro (Preview) (High) | 87.51 |
Interactive version: theaggregate.ai/benchmark?slug=mmmu-pro · How It Works · Data refreshed daily, snapshot 2026-09-05.