ResearchBenchmarksMathematical Reasoning on AIME25 (Final Acc., Correction Uplift)Follow16.67Final AccuracyROSA9.733211.534113.33515.1359Sep 27, 2025Evaluation ResultsMethodMethodLinksFinal AccuracyCorrection UpliftROSAModel=DeepSeek-R1-Dist...Model=DeepSeek-R1-Distill-Qwen-8B2025.0916.6716.67BaselineModel=DeepSeek-R1-Dist...Model=DeepSeek-R1-Distill-Qwen-8B2025.09103.57