Video Reasoning on VideoEspresso extended
63.64Narrative AccuracyTaYS
Evaluation results
| Method | Links | |||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| TaYSBase Model=Qwen2.5-VL-7B-Instruct2026.03 | 63.64 | 35.62 | 28.57 | 25.95 | 46.88 | 35.14 | 41.46 | 34.48 | 51.35 | 16.67 | 47.37 | 66.67 | 46.15 | 10 | 36.86 | |
| Interleaved SFTBase Model=Qwen2.5-VL-7B-Instruct2026.03 | 57.58 | 38.36 | 35.71 | 24.86 | 50 | 35.14 | 48.78 | 24.14 | 37.84 | 16.67 | 36.84 | 44.44 | 46.15 | 20 | 34.98 | |
| Batch w/o thinkingBase Model=Qwen2.5-VL-7B-Instruct2026.03 | 54.55 | 32.88 | 28.57 | 14.59 | 37.5 | 27.03 | 41.46 | 34.48 | 29.73 | 16.67 | 36.84 | 52.94 | 46.15 | 10 | 28.89 | |
| TaYSBase Model=Qwen2.5-VL-3B-Instruct2026.03 | 51.52 | 36.99 | 39.29 | 24.86 | 46.88 | 21.62 | 31.71 | 20.69 | 37.84 | 33.33 | 44.74 | 50 | 53.85 | 20 | 33.45 | |
| Batch SFTBase Model=Qwen2.5-VL-3B-Instruct2026.03 | 48.48 | 34.25 | 17.86 | 17.84 | 43.75 | 32.43 | 39.02 | 27.59 | 24.32 | 8.33 | 39.47 | 50 | 46.15 | 20 | 29.18 | |
| Interleaved SFTBase Model=Qwen2.5-VL-3B-Instruct2026.03 | 48.48 | 38.36 | 25 | 28.65 | 37.5 | 29.73 | 36.59 | 31.03 | 35.14 | 16.67 | 36.84 | 55.56 | 46.15 | 30 | 33.96 | |
| Batch SFTBase Model=Qwen2.5-VL-7B-Instruct2026.03 | 48.48 | 26.03 | 32.14 | 21.62 | 43.75 | 32.43 | 43.9 | 31.03 | 27.03 | 8.33 | 39.47 | 38.89 | 46.15 | 20 | 30.38 | |
| Batch with thinkingBase Model=Qwen2.5-VL-7B-Instruct2026.03 | 42.42 | 31.51 | 42.86 | 23.24 | 37.5 | 29.73 | 29.27 | 37.93 | 29.73 | 25 | 39.47 | 55.56 | 38.46 | 30 | 31.57 | |
| Batch w/o thinkingBase Model=Qwen2.5-VL-3B-Instruct2026.03 | 39.39 | 31.51 | 25 | 18.92 | 37.5 | 24.32 | 43.9 | 20.69 | 29.73 | 16.67 | 42.11 | 38.89 | 30.77 | 10 | 27.99 | |
| Batch with thinkingBase Model=Qwen2.5-VL-3B-Instruct2026.03 | 39.39 | 26.03 | 28.57 | 15.68 | 37.5 | 40.54 | 17.07 | 48.28 | 29.73 | 8.33 | 47.37 | 61.11 | 38.46 | 20 | 28.16 |