@inproceedings{li-etal-2026-syncthink,
    title = "{S}ync{T}hink: A Training-Free Strategy to Align Inference Termination with Reasoning Saturation",
    author = "Li, Gengyang  and
      Cai, Wang  and
      Gao, Yifeng  and
      Wu, Yunfang",
    editor = "Liakata, Maria  and
      Moreira, Viviane P.  and
      Zhang, Jiajun  and
      Jurgens, David",
    booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {ACL} 2026",
    month = jul,
    year = "2026",
    address = "San Diego, California, United States",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl/2026.findings-acl.228/",
    pages = "4657--4672",
    ISBN = "979-8-89176-395-1",
    abstract = "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 tokens attend weakly to early reasoning and focus on `{\ensuremath{<}}/think{\ensuremath{>}}{`}, indicating an information bottleneck.Building on this observation, SyncThink monitors the model{'}s own reasoning-transition signal and terminates reasoning. Experiments on GSM8K, MMLU, GPQA, and BBH across three DeepSeek-R1 distilled models show that SyncThink achieves 62.00{\%} average Top@1 accuracy using 656 generated tokens and 28.68s latency, compared to 61.22{\%}, 2141 tokens, and 92.01s for full CoT decoding. On long-horizon tasks such as GPQA, SyncThink can further yield up to +8.1 absolute accuracy by preventing over-thinking."
}