ResearchBenchmarksMathematical Reasoning on MATH miniFollow36.4EM AccuracyTUR-DPO33.2834.0934.935.71Apr 30, 2026Evaluation ResultsMethodMethodLinksEM AccuracyTUR-DPOEvaluation Protocol=Hu...Evaluation Protocol=Human eval2026.0436.4TUR-DPOEvaluation Protocol=LL...Evaluation Protocol=LLM judge2026.0436PPOEvaluation Protocol=Hu...Evaluation Protocol=Human eval2026.0435.5DPOEvaluation Protocol=Hu...Evaluation Protocol=Human eval2026.0434.1IPOEvaluation Protocol=LL...Evaluation Protocol=LLM judge2026.0433.8DPOEvaluation Protocol=LL...Evaluation Protocol=LLM judge2026.0433.4