ResearchBenchmarksGeneral Capability on GPQA-diamond OpenR1-Math Harder SubsetFollow54AccuracyQwen-4B50.8851.6952.553.31Feb 11, 2026Evaluation ResultsMethodMethodLinksAccuracyQwen-4BBackbone=Qwen, Paramet...Backbone=Qwen, Parameters=4B2026.0254RePOBackbone=Qwen, Paramet...Backbone=Qwen, Parameters=4B2026.0252.5LUFFYBackbone=Qwen, Paramet...Backbone=Qwen, Parameters=4B2026.0251