Strategic Reasoning on VariableSum Dollar OOD (held-out variant)
30.47Win RateDEPT
Evaluation Results
| Method | Links | |
|---|---|---|
| DEPTBackbone=Qwen3-4B-Base2026.05 | 30.47 | |
| GRPOBackbone=Qwen3-4B-Base2026.05 | 27.99 | |
| DEPTBackbone=Qwen3-8B-Base2026.05 | 27.73 | |
| SPAGBackbone=Qwen3-8B-Base2026.05 | 26.56 | |
| SPAGBackbone=Qwen3-4B-Base2026.05 | 26.17 | |
| MARSBackbone=Qwen3-4B-Base2026.05 | 25.39 | |
| MARSBackbone=Qwen3-8B-Base2026.05 | 24.48 | |
| SPIRALBackbone=Qwen3-4B-Base2026.05 | 23.96 | |
| SPIRALBackbone=Qwen3-8B-Base2026.05 | 23.18 | |
| GRPOBackbone=Qwen3-8B-Base2026.05 | 19.14 | |
| VANILLABackbone=Qwen3-4B-Base2026.05 | 3.78 | |
| VANILLABackbone=Qwen3-8B-Base2026.05 | 2.73 |