Video Reasoning on Video-Holmes (Accuracy & Relative Improvement)
52.3AccuracyFFR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| FFRBase Model=Video-R1-SFT, Training=RFT + FFR2026.04 | 52.3 | 51.16 | |
| FFRBase Model=VideoRFT-SFT, Training=RFT + FFR2026.04 | 48 | 77.12 | |
| AVATAR2026.04 | 45.1 | — | |
| Video-Thinker2026.04 | 43.2 | — | |
| GPT-4o2026.04 | 42 | — | |
| VideoRFT2026.04 | 40 | — | |
| Video-R12026.04 | 36.5 | — | |
| Video-ChatR12026.04 | 35.7 | — | |
| Pixel-Reasoner2026.04 | 35.1 | — | |
| Video-R1-SFT2026.04 | 34.6 | — | |
| Qwen2.5-VL2026.04 | 27.8 | — | |
| VideoRFT-SFT2026.04 | 27.1 | — |