ResearchBenchmarksLanguage Understanding on MMLU PRO (ACC. and LEN.)Follow17.58AccuracyGRPO17.403217.449117.49517.5409Mar 19, 2026Evaluation ResultsMethodMethodLinksAccuracyLengthGRPOModel=Qwen3-14bModel=Qwen3-14b2026.0317.581,817DDPOModel=Qwen3-14b, theta...Model=Qwen3-14b, theta=0.22026.0317.411,548