Short Chains, Deep Thoughts: Balancing Reasoning Efficiency and Intra-Segment Capability via Split-Merge Optimization

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.

Large AI models often solve complex problems by generating long, wordy chains of thought, which causes significant delays and high computational costs. While previous methods tried to fix this by simply forcing the AI to use fewer words, this approach often stopped the model from thinking deeply enough to find the right answer. We introduce CoSMo, a new framework that makes AI reasoning faster and smarter by optimizing its structure rather than just cutting its overall word count. CoSMo acts like an automated editor, merging repetitive thoughts and breaking down confusing logical leaps into clear, manageable steps. We then train the model to avoid taking unnecessary reasoning steps, while still allowing it the freedom to use as many words as it needs to thoroughly explain complex ideas within a single step. This structural optimization allows the AI to improve its accuracy on complex tasks while cutting its reasoning steps by nearly 30%.