MOV-Bench - Referential Reasoning: leaderboard
Metric: Accuracy (%; 161 referential-reasoning questions; direct inference on the whole video). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen3 Omni 30B A3B Instruct | 59.63 |
| 2 | MiniCPM-o-4.5 | 58.39 |
| 3 | Ming-Lite-Omni-1.5 | 55.9 |
| 4 | Qwen2.5-Omni-7B | 52.8 |
Interactive version: theaggregate.ai/benchmark?slug=mov-bench-referential-reasoning · How It Works · Data refreshed daily, snapshot 2026-09-26.