Lune

ACL Findings2026

SyncThink: A Training-Free Strategy to Align Inference Termination with Reasoning Saturation

Gengyang Li, Wang Cai, Yifeng Gao, Yunfang Wu

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. Rather than treating reasoning completeness as a theoretically guaranteed criterion, SyncThink uses an empirically calibrated internal stopping signal derived from the model's reasoning-to-answer transition. Our analysis provides mechanistic evidence that answer generation relies disproportionately on the transition token </think>, consistent with an information-bottleneck effect, and we leverage its logit dynamics to detect reasoning saturation and terminate reasoning early. 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 further yields up to +8.1 absolute accuracy by mitigating over-thinking.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Dynamic Early Exit in Reasoning Models cites Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM Reasoning
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on14

  1. Measuring Massive Multitask Language UnderstandingDan Hendrycks, Collin Burns, Steven Basart, Andy Zou et al.ICLR 2021 · 7,905 citations
  2. Let's Verify Step by StepHunter Lightman, Vineet Kosaraju, Yuri Burda, Harrison Edwards et al.ICLR 2024 · 3,045 citations
  3. Efficient Streaming Language Models with Attention SinksGuangxuan Xiao, Yuandong Tian, Beidi Chen, Song Han et al.ICLR 2024 · 1,714 citations
  4. Beyond the 80/20 Rule: High-Entropy Minority Tokens Drive Effective Reinforcement Learning for LLM ReasoningShenzhi Wang, Le Yu, Chang Gao, Chujie Zheng et al.NeurIPS 2025 · 592 citations
  5. Dynamic Early Exit in Reasoning ModelsChenxu Yang, Qingyi Si, Yongjie Duan, Zheliang Zhu et al.ICLR 2026 · 250 citations

Related papers

  1. Stop When Enough: Adaptive Early-Stopping for Chain-of-Thought ReasoningRenliang Sun, Wei Cheng, Dawei Li, Haifeng Chen et al.ACL 2026 · 11 citations
  2. Think Faster Than Words: Efficient LLM Chain-of-Thought Reasoning via Dynamic Shortcut DecodingFan Liu, Yanhao Wang, Min Zhang, Zhikang Chen et al.ACL 2026
  3. When Internalization Fails: Finding Better Targets for Reasoning CompressionMourad Heddaya, Manley Roberts, Rohan Wadhawan, Chenhao TanACL Findings 2026
  4. System-1.5 Reasoning: Traversal in Language and Latent Spaces with Dynamic ShortcutsXiaoqiang Wang, Suyuchen Wang, Yun Zhu, Bang LiuNeurIPS 2025 · 26 citations
  5. Think Better, Not Longer: Token-Level Marginal Utility for Efficient Reasoning in Large Reasoning ModelsJiawei Li, Yang Gao, Huashan Sun, Chong FengACL 2026
Dusk over the sea between two cliffs drawn in fine vertical lines