List your company

Language Agent Task on ScienceWorld (OOD)

29.6ScoreTask-Perturbed NLL Optimization
24.9226.13527.3528.565Jun 25, 2026

Evaluation results

MethodLinks
Task-Perturbed NLL OptimizationBackbone=Qwen3-8B2026.0629.6
Task AugBackbone=Qwen3-8B2026.0626.4
SFTBackbone=Qwen3-8B2026.0625.6
COINBackbone=Qwen3-8B2026.0625.1