ResearchBenchmarksMathematical Reasoning on GSM8K (test) (Accuracy, Training Time, Speedup)Follow93.86AccuracyBPPO44.657657.431370.20582.9787May 27, 2026Evaluation ResultsMethodMethodLinksAccuracyTraining Time (s)Speedup (times)BPPOBase Model=Qwen3-14BBase Model=Qwen3-14B2026.0593.867,0065.22GRPOBase Model=Qwen3-14BBase Model=Qwen3-14B2026.0593.3336,5981Qwen3-14BBase Model=Qwen3-14BBase Model=Qwen3-14B2026.0546.55——