ThinkBrake: A Simple Test-Time Decoding Control for Efficient Reasoning

ACL ARR 2026 January Submission9275 Authors

Keywords: Efficient/Low-Resource Methods for NLP, Language Modeling, Question Answering, AI / LLM Agents

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> token.

Paper Type: Long

Research Area: LLM Efficiency

Research Area Keywords: Efficient/Low-Resource Methods for NLP,Language Modeling,Question Answering,AI / LLM Agents

Contribution Types: Approaches to low-resource settings, Approaches low compute settings-efficiency

Languages Studied: English

Submission Number: 9275