ResearchBenchmarksMathematical Reasoning on AIME OpenR1-Math Harder Subset 2025 (Accuracy)Mathematical Reasoning on AIME OpenR1-Math Harder Subset 2025 (Accuracy)Follow66.1AccuracyRePO45.61250.93156.2561.569Feb 11, 2026Evaluation resultsMethodMethodLinksAccuracyRePOBackbone=Qwen, Paramet...Backbone=Qwen, Parameters=4B2026.0266.1Qwen-4BBackbone=Qwen, Paramet...Backbone=Qwen, Parameters=4B2026.0265.4LUFFYBackbone=Qwen, Paramet...Backbone=Qwen, Parameters=4B2026.0246.4