General Chat Evaluation on Arena-Hard
84Win RateVRM-PPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VRM-PPOBackbone=Qwen 3-8B2026.03 | 84 | 82 | |
| DPOBackbone=Qwen 3-8B2026.03 | 70.8 | 71.9 | |
| KTOBackbone=Qwen 3-8B2026.03 | 69.5 | 70.1 | |
| VRM-PPOBackbone=Qwen2.5-7B2026.03 | 65.5 | 57.2 | |
| IPOBackbone=Qwen 3-8B2026.03 | 65.5 | 54.9 | |
| WPOBackbone=Qwen2.5-7B2026.03 | 62 | 53 | |
| SIMPOBackbone=Qwen 3-8B2026.03 | 61.9 | 56.6 | |
| SELECTIVE DPOBackbone=Qwen 3-8B2026.03 | 60.4 | 51.8 | |
| PPOBackbone=Qwen 3-8B2026.03 | 60.4 | 71.9 | |
| WPOBackbone=Qwen 3-8B2026.03 | 56.2 | 53.7 | |
| PPOBackbone=Qwen2.5-7B2026.03 | 55.3 | 48.9 | |
| IPOBackbone=Qwen2.5-7B2026.03 | 54.6 | 48.3 | |
| SIMPOBackbone=Qwen2.5-7B2026.03 | 54.6 | 48.8 | |
| SELECTIVE DPOBackbone=Qwen2.5-7B2026.03 | 51.7 | 48.2 | |
| DPOBackbone=Qwen2.5-7B2026.03 | 49 | 47.2 | |
| KTOBackbone=Qwen2.5-7B2026.03 | 43.9 | 44.1 |