OThink-R1: Intrinsic Fast/Slow Thinking Mode Switching for Over-Reasoning Mitigation

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhang, Shengjia, Wu, Junjie, Chen, Jiawei, Zhang, Changwang, Li, Zhe, Lou, Xingyu, Zhou, Wangchunshu, Zhou, Sheng, Wang, Can, Wang, Jun
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915710432706560
author Zhang, Shengjia
Wu, Junjie
Chen, Jiawei
Zhang, Changwang
Li, Zhe
Lou, Xingyu
Zhou, Wangchunshu
Zhou, Sheng
Wang, Can
Wang, Jun
author_facet Zhang, Shengjia
Wu, Junjie
Chen, Jiawei
Zhang, Changwang
Li, Zhe
Lou, Xingyu
Zhou, Wangchunshu
Zhou, Sheng
Wang, Can
Wang, Jun
contents Human cognition operates through two complementary modes: fast intuitive thinking and slow deliberate thinking. Vanilla large language models (LLMs) predominantly follow the fast-thinking paradigm, producing immediate responses; while recent large reasoning models (LRMs) adopt slow-thinking strategies, generating detailed reasoning chains before arriving at answers. While LRMs often achieve higher accuracy, this comes at the cost of substantially increased token usage. To address this efficiency-accuracy trade-off, we propose OThink-R1, a hybrid reasoning framework that integrates both modes within a single LRM and enables automatic mode switching based on problem characteristics. We first identify three major patterns of essential and redundant reasoning trajectories in LRMs, which guide the design of an auxiliary LLM-based judge that adaptively determines when slow thinking is necessary. Leveraging the judge's decisions, we construct a hybrid fine-tuning dataset by pruning redundant reasoning to produce fast-thinking samples and retaining complete reasoning for slow-thinking samples. This dataset is then used to fine-tune LRMs, equipping them with inherent autonomous mode-selection capabilities. Extensive experiments on mathematical and question-answering benchmarks show that OThink-R1 reduces reasoning token usage significantly while maintaining competitive accuracy. The code is available at https://github.com/AgenticIR-Lab/OThink-R1.
format Preprint
id arxiv_https___arxiv_org_abs_2506_02397
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle OThink-R1: Intrinsic Fast/Slow Thinking Mode Switching for Over-Reasoning Mitigation
Zhang, Shengjia
Wu, Junjie
Chen, Jiawei
Zhang, Changwang
Li, Zhe
Lou, Xingyu
Zhou, Wangchunshu
Zhou, Sheng
Wang, Can
Wang, Jun
Artificial Intelligence
Human cognition operates through two complementary modes: fast intuitive thinking and slow deliberate thinking. Vanilla large language models (LLMs) predominantly follow the fast-thinking paradigm, producing immediate responses; while recent large reasoning models (LRMs) adopt slow-thinking strategies, generating detailed reasoning chains before arriving at answers. While LRMs often achieve higher accuracy, this comes at the cost of substantially increased token usage. To address this efficiency-accuracy trade-off, we propose OThink-R1, a hybrid reasoning framework that integrates both modes within a single LRM and enables automatic mode switching based on problem characteristics. We first identify three major patterns of essential and redundant reasoning trajectories in LRMs, which guide the design of an auxiliary LLM-based judge that adaptively determines when slow thinking is necessary. Leveraging the judge's decisions, we construct a hybrid fine-tuning dataset by pruning redundant reasoning to produce fast-thinking samples and retaining complete reasoning for slow-thinking samples. This dataset is then used to fine-tune LRMs, equipping them with inherent autonomous mode-selection capabilities. Extensive experiments on mathematical and question-answering benchmarks show that OThink-R1 reduces reasoning token usage significantly while maintaining competitive accuracy. The code is available at https://github.com/AgenticIR-Lab/OThink-R1.
title OThink-R1: Intrinsic Fast/Slow Thinking Mode Switching for Over-Reasoning Mitigation
topic Artificial Intelligence
url https://arxiv.org/abs/2506.02397