Puzzle Solving Agentic Task on Sokoban
89.84Success RateL1
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| L1Backbone Model=Qwen3-4B, Metric Aggregation Type=Best2026.06 | 89.84 | 6.53 | |
| L2Backbone Model=Qwen3-4B, Metric Aggregation Type=Best2026.06 | 88.28 | 6.44 | |
| ReProBackbone Model=Qwen3-4B, Metric Aggregation Type=Best2026.06 | 88.28 | 6.61 | |
| GiGPOBackbone Model=Qwen3-4B, Metric Aggregation Type=Best2026.06 | 86.72 | 6.21 | |
| Meta PromptBackbone Model=Qwen3-4B, Metric Aggregation Type=Best2026.06 | 85.16 | 5.94 | |
| ReProBackbone Model=Qwen3-4B, Metric Aggregation Type=Avg2026.06 | 81.64 | 5.1 | |
| L2Backbone Model=Qwen3-4B, Metric Aggregation Type=Avg2026.06 | 80.08 | 4.79 | |
| L1Backbone Model=Qwen3-4B, Metric Aggregation Type=Avg2026.06 | 79.22 | 4.56 | |
| GiGPOBackbone Model=Qwen3-4B, Metric Aggregation Type=Avg2026.06 | 76.09 | 4.21 | |
| Meta PromptBackbone Model=Qwen3-4B, Metric Aggregation Type=Avg2026.06 | 74.06 | 3.91 |