Lune

ACL Findings2026

ThinkBrake: Efficient Reasoning via Log-Probability Margin Guided Decoding

Sangjun Song, Minjae Oh, Seungkyu Lee, Sungmin Jo, Yohan Jo

Abstract

Large Reasoning Models (LRMs) allocate substantial inference-time compute to Chain-of-Thought (CoT) reasoning, improving performance on mathematics, scientific QA, and tool usage. However, this introduces overthinking: LRMs often reach a correct intermediate solution, continue reasoning, and overwrite it with an incorrect answer. We first demonstrate that oracle stopping-where we inject </think> at every sentence boundary and select the best stopping point in hindsight-improves average accuracy by 8% while reducing thinking tokens by 72%, exposing substantial overthinking. Motivated by this finding, we propose THINKBRAKE, which monitors the log-probability margin between the top continuation token and </think> at sentence boundaries, stopping reasoning when this margin narrows. THINKBRAKE requires no training and achieves favorable accuracy-efficiency trade-offs across math, scientific QA, and tool usage benchmarks, reducing thinking token usage by up to 30%. Furthermore, we provide theoretical analysis showing that THINKBRAKE is equivalent to test-time realignment with a reward bonus for the </think>. Code is available at https: //github.com/holi-lab/ThinkBrake.

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 Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on15

  1. Chain-of-Thought Prompting Elicits Reasoning in Large Language ModelsJason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma et al.NeurIPS 2022 · 22,562 citations
  2. LoRA: Low-Rank Adaptation of Large Language ModelsEdward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu et al.ICLR 2022 · 18,833 citations
  3. Direct Preference Optimization: Your Language Model is Secretly a Reward ModelRafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D. Manning et al.NeurIPS 2023 · 10,924 citations
  4. DAPO: An Open-Source LLM Reinforcement Learning System at ScaleQiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan et al.NeurIPS 2025 · 2,828 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. Let LRMs Break Free from Overthinking via Self-Braking TuningHaoran Zhao, Yuchen Yan, Yongliang Shen, Haolei Xu et al.NeurIPS 2025 · 31 citations
  3. Thinking-Based Non-Thinking: Solving the Reward Hacking Problem in Training Hybrid Reasoning Models via Reinforcement LearningSiyuan Gan, Jiaheng Liu, Boyan Wang, Tianpei Yang et al.ACL 2026
  4. ThinkSwitcher: When to Think Hard, When to Think FastGuosheng Liang, Longguang Zhong, Ziyi Yang, Xiaojun QuanEMNLP Findings 2025 · 1 citation
  5. Think Deep, Not Just Long: Measuring LLM Reasoning Effort via Deep-Thinking TokensWei-Lin Chen, Liqian Peng, Tian Tan, Chao Zhao et al.ICML 2026 · 20 citations
Dusk over the sea between two cliffs drawn in fine vertical lines