Video Preference Modeling on MJBench Video (evaluation)
67.3TauVR-THINKER
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VR-THINKERSize=7B, Protocol=Reasoning-based Reward Models2025.10 | 67.3 | 75.6 | |
| UnifiedRewardSize=7B, Protocol=Generative-based Reward Models2025.10 | 63.3 | 69.5 | |
| UnifiedReward-ThinkSize=7B, Protocol=Reasoning-based Reward Models2025.10 | 62.8 | 71.9 | |
| VideoScoreSize=7B, Protocol=Classifier-based Reward Models2025.10 | 57.9 | 63.5 | |
| VideoRewardSize=2B, Protocol=Classifier-based Reward Models2025.10 | 56.8 | 62.6 | |
| VisionRewardSize=13B, Protocol=Classifier-based Reward Models2025.10 | 54.1 | 65.2 | |
| LiFTSize=13B, Protocol=Generative-based Reward Models2025.10 | 42.5 | 51.4 |