Language Agent Task on ScienceWorld (OOD)
29.6ScoreTask-Perturbed NLL Optimization
Evaluation Results
| Method | Links | |
|---|---|---|
| Task-Perturbed NLL OptimizationBackbone=Qwen3-8B2026.06 | 29.6 | |
| Task AugBackbone=Qwen3-8B2026.06 | 26.4 | |
| SFTBackbone=Qwen3-8B2026.06 | 25.6 | |
| COINBackbone=Qwen3-8B2026.06 | 25.1 |