Video Reasoning on VideoEspresso
58.7AccuracySTORM
Evaluation Results
| Method | Links | |
|---|---|---|
| STORMFrames=32, Backbone=Qwen2.5-VL-7B-Instruct, Sampling Strategy=uniform sampling2026.05 | 58.7 | |
| Video-Thinker-7BFrames=32, Protocol=think with video, Backbone=7B2026.05 | 56.4 | |
| LongVT-RFTFrames=32, Protocol=think with tools2026.05 | 52.7 | |
| Qwen2.5-VL-7B-SFTFrames=32, Backbone=7B, Sampling Strategy=uniform sampling2026.05 | 49.6 | |
| Qwen2.5-VL-7B-Instruct(CoT)Frames=32, Backbone=7B, Sampling Strategy=uniform sampling2026.05 | 45 | |
| Gemini-1.5-ProFrames=322026.05 | 44.2 |