←Back to NewsAI News/ReasoningpaperReasoningPost TrainingRecursive Self-Distillation Doubles Qwen3-8B Math Accuracy to 65.97%A new arXiv paper proposes DCE+SRCL, letting the teacher model co-evolve with its student to lift Qwen3-8B math accuracy by 35 points.SourceAlphaSignalPublishedSep 25, 2026, 12:42 AMAuthorAlphaSignal NewsroomRead1 min readA new arXiv paper proposes DCE+SRCL, letting the teacher model co-evolve with its student to lift Qwen3-8B math accuracy by 35 points.Reporting is indexed from AlphaSignal. Rights remain with the original publisher and cited sources.Read original report ↗Next readsAlphaSignal · paperUCLA's PTTS Gives AI Parallel Reasoning Branches Separate Plans, Gaining 13.4 PointsAlphaSignal · paperLooped Flows Beats Rival Reasoners on ARC-AGI With Just 7M ParametersTinker · newsReViSQL-K2.6 Beats Human Accuracy on SQL Benchmark Without Agentic Scaffolding
