ResearchBenchmarksMathematical Reasoning on AIME 24/25 (dist-3, dist-4, GPT4-Judge)Follow0.2306Dist-3 ScoreGRPO0.2258320.2580160.29020.322384Feb 23, 2026Evaluation ResultsMethodMethodLinksDist-3 ScoreDist-4 ScoreGPT-4 Judge ScoreGRPOBackbone=Qwen2.5-7BBackbone=Qwen2.5-7B2026.020.23060.29022.04KLCovBackbone=Qwen2.5-7BBackbone=Qwen2.5-7B2026.020.24710.32622.07FlowRLBackbone=Qwen2.5-7BBackbone=Qwen2.5-7B2026.020.28780.36542.38EntAdvBackbone=Qwen2.5-7BBackbone=Qwen2.5-7B2026.020.29780.37822.07ClipCovBackbone=Qwen2.5-7BBackbone=Qwen2.5-7B2026.020.33150.41712.1LADBackbone=Qwen2.5-7BBackbone=Qwen2.5-7B2026.020.34980.44422.58