ResearchBenchmarksReasoning on BBH subsetReasoning on BBH subsetFollow47.2AccuracyTUR-DPO43.76844.65945.5546.441Apr 30, 2026Evaluation resultsMethodMethodLinksAccuracyTUR-DPOEvaluation Protocol=Hu...Evaluation Protocol=Human eval2026.0447.2TUR-DPOEvaluation Protocol=LL...Evaluation Protocol=LLM judge2026.0446.7PPOEvaluation Protocol=Hu...Evaluation Protocol=Human eval2026.0446DPOEvaluation Protocol=Hu...Evaluation Protocol=Human eval2026.0444.5IPOEvaluation Protocol=LL...Evaluation Protocol=LLM judge2026.0444.3DPOEvaluation Protocol=LL...Evaluation Protocol=LLM judge2026.0443.9