Interactive Agent on SciWorld (Pass@1, Pass@128)
29.45Pass@1ACE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ACEBackbone=Qwen2.5-7B2026.03 | 29.45 | 59.67 | |
| LEAFEBackbone=Qwen2.5-7B2026.03 | 27.88 | 62 | |
| GRPO-RLVRBackbone=Qwen2.5-7B2026.03 | 27.17 | 57.33 | |
| EARLY-EXPBackbone=Qwen2.5-7B2026.03 | 26.17 | 54.67 | |
| ACEBackbone=Llama3.1-8B2026.03 | 25.28 | 57.33 | |
| GRPO-RLVRBackbone=Llama3.1-8B2026.03 | 24.25 | 56 | |
| EARLY-EXPBackbone=Llama3.1-8B2026.03 | 24.04 | 56 | |
| LEAFEBackbone=Llama3.1-8B2026.03 | 22.7 | 59.33 | |
| BASE (NO FT)Backbone=Llama3.1-8B2026.03 | 7.17 | 48.67 | |
| BASE (NO FT)Backbone=Qwen2.5-7B2026.03 | 7 | 47.33 |