Textual Environment Interaction on ScienceWorld
96.06Base ScoreQWEN-2.5-3B-SFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| QWEN-2.5-3B-SFTBackbone=Qwen-2.5, Size=3B, Mode=SFT2026.02 | 96.06 | 89.73 | |
| QWEN-2.5-3B-SWIRLBackbone=Qwen-2.5, Size=3B, Mode=SWIRL2026.02 | 96.06 | 89.92 | |
| QWEN-2.5-3B-INSTRUCTBackbone=Qwen-2.5, Size=3B, Mode=Instruct2026.02 | 82.87 | 18.68 | |
| DEEPSEEK-7B-CHATBackbone=DeepSeek, Size=7B, Mode=Chat2026.02 | 81.9 | 16.35 | |
| QWEN-2.5-14B-INSTRUCTBackbone=Qwen-2.5, Size=14B, Mode=Instruct2026.02 | 81.46 | 15.52 | |
| QWEN-2.5-32B-INSTRUCTBackbone=Qwen-2.5, Size=32B, Mode=Instruct2026.02 | 81.01 | 14.53 | |
| QWEN-2.5-7B-INSTRUCTBackbone=Qwen-2.5, Size=7B, Mode=Instruct2026.02 | 80.48 | 12.5 | |
| OLMO-3-7B-INSTRUCTBackbone=OLMo-3, Size=7B, Mode=Instruct2026.02 | 79.27 | 11.28 |