arXiv:2602.03141v4 Announce Type: replace 
Abstract: While Large Reasoning Models (LRMs) have demonstrated impressive capabilities in solving complex tasks through the generation of long reasoning chains, this reliance on verbose generation results in significant latency and computational overhead. To address these challenges, we propose \textbf{CoSMo} (\textbf{Co}nsistency-Guided \textbf{S}plit-\textbf{M}erge \textbf{O}ptimization), a framework designed to eliminate structural redundancy rather than indiscriminately restricting token volume. Specifically, CoSMo utilizes a split-merge algorithm that dynamically refines reasoning chains by merging redundant segments and splitting logical gaps to ensure coherence. We then employ structure-aligned reinforcement learning with a novel segment-level budget to supervise the model in maintaining efficient reasoning structures throughout training. Extensive experiments across multiple benchmarks and backbones demonstrate that CoSMo achieves superior performance, improving accuracy by \textbf{3.3} points while reducing segment usage by \textbf{28.7\%} on average compared to reasoning efficiency baselines.

أدت التطورات الأخيرة في نماذج التفكير الكبيرة (LRMs) إلى تطوير CoSMo، وهو إطار يعمل على تحسين كفاءة التفكير من خلال القضاء على التكرارات الهيكلية في سلاسل التفكير. تستخدم هذه الطريقة خوارزمية تقسيم-دمج لتنقيح القطاعات المنطقية، مما يعزز التماسك ويقلل من الحمل الحسابي.

Los avances recientes en los Modelos de Razonamiento Grande (LRMs) han llevado al desarrollo de CoSMo, un marco que optimiza la eficiencia del razonamiento al eliminar redundancias estructurales en las cadenas de razonamiento. Este enfoque utiliza un algoritmo de división-fusión para refinar segmentos lógicos, mejorando la coherencia y reduciendo la sobrecarga computacional.

Des avancées récentes dans les grands modèles de raisonnement (LRMs) ont conduit au développement de CoSMo, un cadre qui optimise l'efficacité du raisonnement en éliminant les redondances structurelles dans les chaînes de raisonnement. Cette approche utilise un algorithme de séparation-fusion pour affiner les segments logiques, améliorant la cohérence tout en réduisant la surcharge computationnelle.

Recent advancements in Large Reasoning Models (LRMs) have led to the development of CoSMo, a framework that optimizes reasoning efficiency by eliminating structural redundancies in reasoning chains. This approach utilizes a split-merge algorithm to refine logical segments, enhancing coherence while reducing computational overhead.