ResearchBenchmarksDecision Inference on SMACFollow76.4AccuracyOurs44.26452.60760.9569.293Feb 18, 2025Evaluation ResultsMethodMethodLinksAccuracyOurs2025.0276.4ReFTFramework=ReasoningFramework=Reasoning2025.0272.2KTOFramework=RLHFFramework=RLHF2025.0272.1DPOFramework=RLHFFramework=RLHF2025.0271.3SkyworkFramework=RLAIFFramework=RLAIF2025.0269.2DeepSeek2025.0265.8PPOFramework=RLHFFramework=RLHF2025.0265.3CoT+SFTFramework=ReasoningFramework=Reasoning2025.0264.2Proxy LLM2025.0264SFT2025.0258.2o3-mini2025.0245.5