Preference Prediction on VideoGen-RewardBench
76.81Score @ Threshold 99PRISM
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| PRISMTie handling=w/o Ties, Backbone=Wan2.1-14B2026.06 | 76.81 | 76.44 | 75.78 | 74.79 | 72.63 | 69.87 | 60.36 | 74.12 | |
| UnifiedRewardTie handling=w/o Ties2026.06 | 75.83 | 63.72 | 52.99 | 49.37 | 50.19 | 50.3 | 49.86 | 77.38 | |
| PRISMTie handling=w/o Ties, Backbone=Wan2.1-1.3B2026.06 | 74.51 | 74.42 | 73.43 | 72.68 | 70.01 | 67.56 | 59.56 | 72.93 | |
| VideoRewardTie handling=w/o Ties2026.06 | 67.81 | 59.88 | 51.87 | 52.55 | 51.84 | 51.74 | 50.64 | 72.38 | |
| PRISMTie handling=w/ Ties, Backbone=Wan2.1-14B2026.06 | 63.98 | 63.7 | 63.13 | 62.3 | 60.46 | 58.16 | 50.25 | 61.68 | |
| PRISMTie handling=w/o Ties, Backbone=CogVideoX-2B2026.06 | 63.19 | 63.08 | 62.89 | 62.64 | 60.31 | 58.24 | 51.5 | 61.8 | |
| UnifiedRewardTie handling=w/ Ties2026.06 | 63.1 | 53.02 | 44.17 | 41.6 | 42.19 | 42.3 | 41.93 | 64.39 | |
| PRISMTie handling=w/ Ties, Backbone=Wan2.1-1.3B2026.06 | 62.07 | 61.99 | 61.13 | 60.5 | 58.28 | 56.26 | 49.6 | 60.76 | |
| VideoScore2Tie handling=w/o Ties2026.06 | 56.76 | 52.89 | 26.49 | 16.86 | 17.87 | 17.82 | 17.2 | 58.27 | |
| VideoRewardTie handling=w/ Ties2026.06 | 56.43 | 49.83 | 43.16 | 43.73 | 43.14 | 43.05 | 42.13 | 60.23 | |
| PRISMTie handling=w/ Ties, Backbone=CogVideoX-2B2026.06 | 52.61 | 52.53 | 52.36 | 52.15 | 50.25 | 48.54 | 43.25 | 51.44 | |
| VideoScore2Tie handling=w/ Ties2026.06 | 48.03 | 44.36 | 27.19 | 20.68 | 21.17 | 21.31 | 21.02 | 49.23 |