MOV-Bench - Intent Reasoning: leaderboard
Metric: Accuracy (%; 80 intent-reasoning questions; direct inference on the whole video). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen3 Omni 30B A3B Instruct | 55 |
| 2 | Ming-Lite-Omni-1.5 | 50 |
| 3 | MiniCPM-o-4.5 | 48.75 |
| 4 | Qwen2.5-Omni-7B | 42.5 |
Interactive version: theaggregate.ai/benchmark?slug=mov-bench-intent-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-26.