Open-ended Generation on Arena-Hard
84.6ScoreAR-MAP
Evaluation Results
| Method | Links | |
|---|---|---|
| AR-MAPBackbone=Dream-7B-Instruct, Alignment Method=AR-MAP2026.02 | 84.6 | |
| VRPOBackbone=Dream-7B-Instruct, Alignment Method=VRPO2026.02 | 81.8 | |
| AR-MAPBackbone=SDAR-8B-Instruct, Alignment Method=AR-MAP2026.02 | 73.55 | |
| SimPOBackbone=Dream-7B-Instruct, Alignment Method=SimPO2026.02 | 72.2 | |
| DPOBackbone=Dream-7B-Instruct, Alignment Method=DPO2026.02 | 71.23 | |
| VRPOBackbone=SDAR-8B-Instruct, Alignment Method=VRPO2026.02 | 68.14 | |
| DPOBackbone=Qwen2.5-7B, Alignment Method=DPO2026.02 | 67.8 | |
| DPOBackbone=SDAR-8B-Instruct, Alignment Method=DPO2026.02 | 61.12 | |
| SimPOBackbone=SDAR-8B-Instruct, Alignment Method=SimPO2026.02 | 61.08 | |
| DPOBackbone=Qwen3-8B-Base, Alignment Method=DPO2026.02 | 60.4 | |
| Qwen3-8B-BaseBackbone=Qwen3-8B-Base, Alignment Method=Base2026.02 | 57.4 | |
| Dream-7B-InstructBackbone=Dream-7B-Instruct, Alignment Method=Instruct2026.02 | 55.04 | |
| Qwen2.5-7BBackbone=Qwen2.5-7B, Alignment Method=Base2026.02 | 50.2 | |
| SDAR-8B-InstructBackbone=SDAR-8B-Instruct, Alignment Method=Instruct2026.02 | 41.08 |