ResearchBenchmarksMathematical Reasoning on MATH-500 OpenR1 HarderFollow96.6AccuracyQwen-4B93.68894.44495.295.956Feb 11, 2026Evaluation ResultsMethodMethodLinksAccuracyQwen-4BBackbone=Qwen, Paramet...Backbone=Qwen, Parameters=4B2026.0296.6RePOBackbone=Qwen, Paramet...Backbone=Qwen, Parameters=4B2026.0296.6LUFFYBackbone=Qwen, Paramet...Backbone=Qwen, Parameters=4B2026.0293.8