Strategic Reasoning on RandomValue Negotiation OOD (held-out variant)
17.08Win RateDEPT
Evaluation Results
| Method | Links | |
|---|---|---|
| DEPTBackbone=Qwen3-8B-Base2026.05 | 17.08 | |
| DEPTBackbone=Qwen3-4B-Base2026.05 | 15.36 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 14.52 | |
| SPAGBackbone=Qwen3-8B-Base2026.05 | 13.02 | |
| SPIRALBackbone=Qwen3-4B-Base2026.05 | 12.56 | |
| MARSBackbone=Qwen3-4B-Base2026.05 | 12.55 | |
| SPAGBackbone=Qwen3-4B-Base2026.05 | 11.85 | |
| MARSBackbone=Qwen3-8B-Base2026.05 | 7.56 | |
| GRPOBackbone=Qwen3-8B-Base2026.05 | 6.64 | |
| SPIRALBackbone=Qwen3-8B-Base2026.05 | 6.52 | |
| VANILLABackbone=Qwen3-8B-Base2026.05 | 3.78 | |
| VANILLABackbone=Qwen3-4B-Base2026.05 | 0.13 |