ResearchBenchmarksDecision Inference on AI2-THORFollow70.2Success RateOurs43.6850.56557.4564.335Feb 18, 2025Evaluation ResultsMethodMethodLinksSuccess RateOurs2025.0270.2o3-mini2025.0267.7ReFTFramework=ReasoningFramework=Reasoning2025.0264.2KTOFramework=RLHFFramework=RLHF2025.0262.8SkyworkFramework=RLAIFFramework=RLAIF2025.0248.3DeepSeek2025.0247.6SFT2025.0244.7