Video-MMMU — leaderboard
300 college-level lecture videos across 30 subjects. Measures whether watching video helps LLMs solve problems - the 'delta-knowledge' metric. Humans gain 33% from video; top models only 15%.
Metric: Overall Score. Source: videommmu.github.io. Status: saturated. 25 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Human Expert | 66 |
| 2 | Claude 3.5 Sonnet | 65.78 |
| 3 | GPT-4o | 61.22 |
| 4 | Gemini 1.5 Pro | 53.89 |
| 5 | Gemini 1.5 Flash | 49.78 |
| 6 | InternVL2-8B | 37.44 |
Interactive version: theaggregate.ai/benchmark?slug=video-mmmu · How the rankings work · Data refreshed daily, snapshot 2026-07-22.