ResearchBenchmarksExplanation Generation on VariErrFollow0.0839SFT (Mean Feature KL)Qwen0.0827320.0906160.09850.106384May 27, 2026Evaluation ResultsMethodMethodLinksSFT (Mean Feature KL)CAPO (Mean Feature KL)Relative ChangeQwenModel=QwenModel=Qwen2026.050.08390.08064LlamaModel=LlamaModel=Llama2026.050.11310.12147.3