Human Preference Prediction on internal benchmark
82.2AccuracyQwen-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen-7BConfiguration=GCPO2026.04 | 82.2 | |
| Seed-1.5-VLConfiguration=T+V (Think+Verify)2026.04 | 79.3 | |
| Seed-1.6-VLConfiguration=T+V (Think+Verify)2026.04 | 77.2 | |
| Qwen-7BConfiguration=T+V (Think+Verify)2026.04 | 75.4 | |
| Seed-1.5-VLConfiguration=T (Think)2026.04 | 72.2 | |
| Qwen-3BConfiguration=GCPO2026.04 | 72 | |
| Seed-1.6-VLConfiguration=T (Think)2026.04 | 71.2 | |
| Qwen-7BConfiguration=V (Verify)2026.04 | 70.9 | |
| Seed-1.6-VLConfiguration=V (Verify)2026.04 | 69.4 | |
| Qwen-3BConfiguration=T+V (Think+Verify)2026.04 | 69.3 | |
| Qwen-7BConfiguration=T (Think)2026.04 | 68.9 | |
| Qwen-7B (VIESCORE)Configuration=V (Verify)2026.04 | 68.3 | |
| Qwen-3BConfiguration=V (Verify)2026.04 | 66.1 | |
| Qwen-3BConfiguration=T (Think)2026.04 | 64.1 |