Pairwise Preference on GenAI Bench (test)
76.3AccuracyHPSv3++
Evaluation Results
| Method | Links | |
|---|---|---|
| HPSv3++2026.06 | 76.3 | |
| UnifiedReward2026.06 | 72.4 | |
| UnifiedRewardBackbone=LLaVA-OV-7B2026.02 | 72.38 | |
| HPSv3Backbone=Qwen2VL-7B2026.02 | 70.95 | |
| UnifiedReward-ThinkBackbone=LLaVA-OV-7B2026.02 | 70.91 | |
| HPSv32026.06 | 70.8 | |
| PickScore2026.06 | 70 | |
| PickScoreBackbone=CLIP2026.02 | 69.98 | |
| DiNa-LRMBackbone=SD3.5-M-2B, Multi-noise inference scaling=true2026.02 | 68.98 | |
| DiNa-LRMBackbone=SD3.5-M-2B, Multi-noise inference scaling=false2026.02 | 68.43 | |
| MPSBackbone=CLIP2026.02 | 68.08 | |
| HPSv2Backbone=CLIP2026.02 | 67.62 | |
| ImageRewardBackbone=CLIP2026.02 | 63.41 | |
| ImageReward2026.06 | 63.4 | |
| LRM-SDXLBackbone=SDXL2026.02 | 61.58 | |
| LRM-SD1.5Backbone=SD-1.52026.02 | 60.86 | |
| Aesthetic Score2026.06 | 57.3 | |
| CLIP ViT-H/142026.06 | 56 |