Text-to-image preference evaluation on HPD v3 (test)
78.3AccuracyGemini 3.1 Pro
Evaluation results
| Method | Links | |
|---|---|---|
| Gemini 3.1 ProARR=true2026.05 | 78.3 | |
| HPSv32026.05 | 76.9 | |
| Gemini 3.1 ProARR=false2026.05 | 76.6 | |
| GPT-5ARR=true2026.05 | 76.1 | |
| GPT-5ARR=false2026.05 | 72.4 | |
| Qwen3-VL-8BARR=true2026.05 | 70.2 | |
| UnifiedReward-Thinking2026.05 | 68.1 | |
| Qwen3-VL-8BARR=false2026.05 | 67.2 | |
| UnifiedReward2026.05 | 66 | |
| PickScore2026.05 | 65.6 | |
| ImageReward2026.05 | 58.6 |