List your company

Text-based Science Agent Tasks on ScienceWorld

35AccuracyDPO+
24.627.33032.7Jun 25, 2026

Evaluation results

MethodLinks
DPO+Backbone=Qwen3-8B2026.0635
FRODOBackbone=Qwen3-8B2026.0630
DPOBackbone=Qwen3-8B2026.0629
SFT+Backbone=Qwen3-8B2026.0627
SFTBackbone=Qwen3-8B2026.0625