一作ACL 2026Findings· Findings of ACL
S ync T hink: A Training-Free Strategy to Align Inference Termination with Reasoning Saturation
Gengyang Li · Wang Cai · Yifeng Gao · Yunfang Wu
摘要
Chain-of-Thought (CoT) prompting improves reasoning but often produces long and redundant traces that substantially increase inference cost. We present SyncThink, a training-free and plug-and-play decoding method that reduces CoT overhead without modifying model weights. We find that answer token...