Sequential Decision Making on ScienceWorld
70.4Pass@1 Success RateAutoRefine
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| AutoRefineBackbone=GPT-4-turbo2026.01 | 70.4 | 16.5 | |
| ReAct + ReflexionBackbone=GPT-4-turbo2026.01 | 69.2 | 40.2 | |
| ReflexionBackbone=GPT-4-turbo2026.01 | 67.4 | 42 | |
| ReActBackbone=GPT-4-turbo2026.01 | 61.8 | 26.2 |