Demystifying Hybrid Thinking: Can LLMs Truly Switch Between Think and No-Think?

2025/10/14 by Shouren Wang, Yang Wang, Wang, Shouren +9 · 1 citation
Business, Management and Accounting · Computer Science · #Artificial Intelligence (cs.AI) #Computation and Language (cs.CL) #FOS: Computer and information sciences #FinTech, Crowdfunding, Digital Finance #Machine Learning (cs.LG) #Open Education and E-Learning

paper · pdf · doi:10.48550/arxiv.2510.12680

openalex publication_date 2025/10/14 · openalex created_date 2025/10/17 · openalex updated_date 2026/07/28

Abstract

Hybrid thinking enables LLMs to switch between reasoning and direct answering, offering a balance between efficiency and reasoning capability. Yet our experiments reveal that current hybrid thinking LLMs only achieve partial mode separation: reasoning behaviors often leak into the no-think mode. To understand and mitigate this, we analyze the factors influencing controllability and identify four that matter most: (1) larger data scale, (2) using think and no-think answers from different questions rather than the same question, (3) a moderate increase in no-think data number, and (4) a two-phase strategy that first trains reasoning ability and then applies hybrid think training. Building on these findings, we propose a practical recipe that, compared to standard training, can maintain accuracy in both modes while significantly reducing no-think output length (from 1085 to 585 on MATH500) and occurrences of reasoning-supportive tokens such as ``wait'' (from 5917 to 522 on MATH500). Our findings highlight the limitations of current hybrid thinking and offer directions for strengthening its controllability.

Citations

• gpt-oss-120b & gpt-oss-20b Model Card
• Qwen3 Technical Report
• Understanding R1-Zero-Like Training: A Critical Perspective
• Gemma 3 Technical Report
• Think Twice: Enhancing LLM Reasoning by Scaling Multi-round Test-time Thinking
• SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wild
• Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models
• DAPO: An Open-Source LLM Reinforcement Learning System at Scale
• Thinking Machines: A Survey of LLM based Reasoning Strategies
• Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Models
• Unveiling the Key Factors for Distilling Chain-of-Thought Reasoning
• From System 1 to System 2: A Survey of Reasoning Large Language Models
• LightThinker: Thinking Step-by-Step Compression
• TokenSkip: Controllable Chain-of-Thought Compression in LLMs
• LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!
• Demystifying Long Chain-of-Thought Reasoning in LLMs
• s1: Simple test-time scaling
• Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs
• Kimi k1.5: Scaling Reinforcement Learning with LLMs
• DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning
• Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs
• Qwen2.5 Technical Report
• Unlocking the Capabilities of Thought: A Reasoning Boundary Framework to Quantify and Optimize Chain-of-Thought
• DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Model
• BioXP-0.5B: Explainable Medical-AI via RL-GRPO
• A Survey of Reasoning with Foundation Models
• GPQA: A Graduate-Level Google-Proof Q&A Benchmark
• Let's Verify Step by Step
• Proximal Policy Optimization Algorithms
• OpenAI o1 System Card

Cited by

• Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Claude Code Agent Teams

Related

• Think Only When You Need with Large Hybrid-Reasoning Models
• Thinkless: LLM Learns When to Think
• Thinking LLMs: General Instruction Following with Thought Generation
• Reasoning Models Can Be Effective Without Thinking
• Don't Think Longer, Think Wisely: Optimizing Thinking Dynamics for Large Reasoning Models