MMVU: leaderboard
3,000 expert-written questions over 1,529 specialist videos across 27 subjects in science, healthcare, humanities and engineering (Yale NLP, 2025); 1,858 multiple-choice and 1,142 open-ended items.
Metric: Hidden-test overall accuracy, chain-of-thought (%). Source: arxiv.org. Source: mmvu-benchmark.github.io. Saturation forecast: Estimated already saturated. 36 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O1 | 76.1 | #183 |
| 2 | Gemini 2.0 Flash (Thinking) | 69.5 | #342 (Gemini 2.0 Flash) |
| 3 | GPT-4o | 66.7 | #343 |
| 4 | Gemini 2.0 Flash | 66.5 | #342 |
| 5 | Gemini 1.5 Pro | 65.8 | #445 |
| 6 | Claude 3.5 Sonnet | 64.1 | #350 |
| 7 | Grok 2 | 63.4 | #452 |
| 8 | GPT-4o Mini | 61.5 | #599 |
| 9 | Gemini 1.5 Flash | 58.8 | #618 |
| 10 | Qwen 2 VL 72B | 53.2 | #460 |
| 11 | Qwen 2 VL 7B Instruct | 42.5 | #826 |
| 12 | Qwen 2 VL 2B | 36.5 | #1220 |
| 13 | InternVL2-8B | 36.2 | #817 |
| 14 | Pixtral-12B | 32.2 | #808 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmvu · How It Works · Data refreshed daily, snapshot 2026-10-20.