arXiv cs.AI· Weixin Guan, Liang Li, Shiyu Xie, Jiapeng Liu, Bing Li, Peng Fu, Chengyang Fang, Xiaoshuai Hao, Can Ma, Weiping Wang·· 6 天前AI 评分22
推理模型过度思考的最优终止方法 Pareto 前沿构建
Constructing Pareto Frontiers of Optimal Reasoning Termination Methods for Overthinking in Reasoning Models
AI 导读
研究提出 Step-wise Truncation Probing（STP），沿固定采样推理轨迹反复从多个保留前缀生成答案，并据此定义 Overthinking Ratio 与 Accuracy Degradation，用于刻画模型-任务层面的过度思考程度。
正文 · 原文
Abstract:Reasoning models can benefit from test-time scaling, but continued reasoning may become ineffective or harmful, a phenomenon termed overthinking. Evaluations against full thinking compare accuracy and token usage but lack references for overthinking severity and strategies' theoretical mitigation potential. To characterize overthinking at the model-task level, we first introduce Step-wise Truncation Probing (STP), repeatedly generating answers from multiple retained prefixes along each fixed sampled reasoning trajectory. Using each trajectory's earliest measured position attaining maximum empirical accuracy as a reference, we define Overthinking Ratio and Accuracy Degradation as the fraction of thinking tokens generated afterward and the accuracy decrease from this reference to full thinking, respectively. Secondly, we then formulate dynamic routing, fixed budgets, and dynamic early exit within a unified framework to quantify the theoretical mitigation potential of different termination strategies. Based on this unified framework, we use the complete STP measurements to construct a token-accuracy Pareto frontier for each strategy under a common normalized thinking-token budget. We further develop OTBench, a reusable benchmark built from the STP process, which produced 70.5B reasoning tokens and the corresponding probing measurements across six models and four benchmark groups.
| Subjects: | Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.13570 [cs.AI] |
| (or arXiv:2610.13570v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2610.13570 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Liang Li [view email]
[v1]
Fri, 9 Oct 2026 13:22:37 UTC (2,379 KB)
来源：arXiv cs.AI · arxiv.org