mlsanityData to 7 Oct 2026

ThinkBrake: Efficient Reasoning via Log-Probability Margin Guided Decoding

arXiv 2510.00546 ↗First posted 1 Oct 2025cs.CLcited by 0

On the map: LLM reasoning → Test-time compute and efficient thinking

Abstract (arXiv)

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> token.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 16 of its 41 references that are in the map. First posted on arXiv 1 Oct 2025.

Builds on

16 of 41 references.

Closest ancestors · 4

Same direction, newest first

  1. 2025ThinkLess: A Training-Free Inference-Efficient Method for Reducing Reasoning Redundancy21 May
  2. 2025FlashThink: An Early Exit Method For Efficient Reasoning20 May
  3. 2025Concise Reasoning via Reinforcement Learning7 Apr
  4. 2024C3oT: Generating Shorter Chain-of-Thought without Compromising Effectiveness16 Dec

Other references · 12

  1. 2021Training Verifiers to Solve Math Word Problemscited by 9,743
  2. 2025Qwen3 Technical Reportcited by 8,195
  3. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  4. 2025DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learningcited by 6,146
  5. 2018Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challengecited by 4,710
  6. 2024Phi-4 Technical Reportcited by 675
  7. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Modelscited by 381
  8. 2025Towards Large Reasoning Models: A Survey of Reinforced Reasoning with Large Language Modelscited by 291
  9. 2024Omni-MATH: A Universal Olympiad Level Mathematic Benchmark For Large Language Modelscited by 277
  10. 2022On Reinforcement Learning and Distribution Matching for Fine-Tuning Language Models with no Catastrophic Forgettingcited by 80
  11. 2025Early Stopping Chain-of-thoughts in Large Language Modelscited by 19
  12. 2025Walk Before You Run! Concise LLM Reasoning via Reinforcement Learningcited by 15

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Logit Arithmetic Elicits Long Reasoning Capabilities Without Training
  2. 2026TERMINATOR: Learning Optimal Exit Points for Early Stopping in Chain-of-Thought Reasoning
  3. 2025THINK-Bench: Evaluating Thinking Efficiency and Chain-of-Thought Quality of Large Reasoning Models
  4. 2026SyncThink: A Training-Free Strategy to Align Inference Termination with Reasoning Saturation
  5. 2025Speculative Thinking: Enhancing Small-Model Reasoning with Large Model Guidance at Inference Time
  6. 2026When More Thinking Hurts: Overthinking in LLM Test-Time Compute Scaling
  7. 2025Done Is Better than Perfect: Unlocking Efficient Reasoning by Structured Multi-Turn Decomposition
  8. 2026Efficient Reasoning via Constrained Optimization in Latent Space