Video Question Answering on MVBench (Accuracy, Think Ratio, Response Length)
72AccuracyVideoAuto-R1
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| VideoAuto-R1Reasoning Mode=AutoThink, Backbone=Qwen3-VL-8B2026.01 | 72 | 31 | 52 | |
| VideoAuto-R1Reasoning Mode=AutoThink, Backbone=Qwen2.5-VL-7B2026.01 | 71 | 25 | 44 | |
| VideoChat-R1.5Reasoning Mode=Think-Only2026.01 | 70.6 | — | 133 | |
| Qwen3-VL-8BReasoning Mode=None, Reproduced=true2026.01 | 69.4 | — | 2.2 | |
| Temporal-RLTReasoning Mode=Think-Only2026.01 | 68.1 | — | — | |
| LongVILA-R1Reasoning Mode=Think-Only2026.01 | 67.6 | — | — | |
| Qwen2.5-VL-7BReasoning Mode=None, Reproduced=true2026.01 | 67.1 | — | 3 | |
| LOVE-R1Reasoning Mode=Think-Only2026.01 | 66.6 | — | — | |
| Video-R1Reasoning Mode=Think-Only2026.01 | 65.5 | — | 386 | |
| Video-RFTReasoning Mode=Think-Only2026.01 | 62.1 | — | — |