Multi-discipline Video Understanding on MMVU
75.4AccuracyGPT-4o
Evaluation Results
| Method | Links | |
|---|---|---|
| GPT-4oFrames=322026.05 | 75.4 | |
| STORMFrames=32, Backbone=Qwen2.5-VL-7B-Instruct, Sampling Strategy=uniform sampling2026.05 | 65.9 | |
| Video-R1-7BFrames=32, Backbone=7B, Sampling Strategy=uniform sampling2026.05 | 64.2 | |
| Video-Thinker-7BFrames=32, Protocol=think with video, Backbone=7B2026.05 | 63.4 | |
| Qwen2.5-VL-7B-SFTFrames=32, Backbone=7B, Sampling Strategy=uniform sampling2026.05 | 62.1 | |
| LongVT-RFTFrames=32, Protocol=think with tools2026.05 | 61.4 | |
| Qwen2.5-VL-7B-Instruct(CoT)Frames=32, Backbone=7B, Sampling Strategy=uniform sampling2026.05 | 60 |