Agent Task on Minesweeper
50.7Suc@1SKillRL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SKillRLBackbone=Qwen3-4B2026.07 | 50.7 | 56.3 | 57.9 | |
| PivoARLBackbone=Qwen3-4B2026.07 | 48.4 | 71.8 | 78.1 | |
| GSPOBackbone=Qwen3-4B2026.07 | 46.1 | 57 | 57.8 | |
| DAPOBackbone=Qwen3-4B2026.07 | 46 | 50 | 51.6 | |
| Mem0+GRPOBackbone=Qwen3-4B2026.07 | 44.5 | 46.8 | 49.2 | |
| SimpleMem+GRPOBackbone=Qwen3-4B2026.07 | 43 | 50 | 51.5 | |
| GiGPOBackbone=Qwen3-4B2026.07 | 42.2 | 46.1 | 48.4 | |
| MetaRLBackbone=Qwen3-4B2026.07 | 39.8 | 54.9 | 61.5 | |
| GRPOBackbone=Qwen3-4B2026.07 | 39.1 | 42.1 | 45.3 | |
| Gemini-2.5-ProBackbone=Gemini-2.5-Pro2026.07 | 24.2 | 29.6 | 31.2 | |
| GPT-4oBackbone=GPT-4o2026.07 | 22.5 | 26.4 | 26.7 | |
| ReActBackbone=Qwen3-4B2026.07 | 3.12 | 3.96 | 5.83 | |
| ReflexionBackbone=Qwen3-4B2026.07 | 3.12 | 5.46 | 7.49 |