MMOU: leaderboard
NVIDIA omni-understanding benchmark (Massive Multi-task Omni Understanding): 20,000 curated multiple-choice questions over 11,877 web videos averaging 9.6 minutes across 10 domains, 35 subcategories and 13 skills, where the answer requires joint reasoning over visual, audio and textual signals rather than any one channel. Human reference 84.3%.
Metric: Accuracy (%). Source: mmou.pages.dev. Saturation forecast: Around February 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 82.67 | #64 |
| 2 | Gemini 3.6 Flash | 82.5 | #57 |
| 3 | Gemini 3.7 Flash | 82.5 | #24 |
| 4 | Gemini 3.5 Flash | 81.9 | #66 |
| 5 | Gemini 2.5 Pro | 64.2 | #159 |
| 6 | Gemini 2.5 Flash | 55.8 | #255 |
| 7 | Qwen3 Omni 30B A3B Instruct | 46 | #401 |
| 8 | Qwen 3 VL 32B Instruct | 41.21 | #290 |
| 9 | Qwen 3 VL 8B Instruct | 35.63 | #419 |
| 10 | GPT-4.1 | 33.9 | #258 |
| 11 | Qwen 3 235B A22B | 32.73 | #323 |
| 12 | Qwen2.5-Omni-7B | 29.81 | #633 |
| 13 | Qwen 2.5 VL 7B Instruct | 28.36 | #650 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mmou · How It Works · Data refreshed daily, snapshot 2026-11-05.