mlsanityData to 7 Oct 2026

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

arXiv 2510.12680 ↗First posted 14 Oct 2025cs.LG cs.AI cs.CLcited by 9

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

Abstract (arXiv)

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. The code is available at: https://github.com/SR-A-W/demystifying-hybrid-thinking

Key facts. Cited by 9 papers in the map (data through 7 Oct 2026). Placed in Test-time compute and efficient thinking, one of 17 directions in LLM reasoning. Builds on 40 of its 63 references that are in the map. First posted on arXiv 14 Oct 2025.

Builds on

40 of 63 references.

Closest ancestors · 13

Same direction, newest first

  1. 2025gpt-oss-120b & gpt-oss-20b Model Card8 Aug
  2. 2025Speculative Thinking: Enhancing Small-Model Reasoning with Large Model Guidance at Inference Time12 Apr
  3. 2025Think Twice: Enhancing LLM Reasoning by Scaling Multi-round Test-time Thinking25 Mar
  4. 2025Stop Overthinking: A Survey on Efficient Reasoning for Large Language Models20 Mar
  5. 2025The First Few Tokens Are All You Need: An Efficient and Effective Unsupervised Prefix Fine-Tuning Method for Reasoning Models4 Mar
  6. 2025How Well do LLMs Compress Their Own Chain-of-Thought? A Token Complexity Approach3 Mar
  7. 2025LightThinker: Thinking Step-by-Step Compression21 Feb
  8. 2025TokenSkip: Controllable Chain-of-Thought Compression in LLMs17 Feb
  9. 2025The Danger of Overthinking: Examining the Reasoning-Action Dilemma in Agentic Tasks12 Feb
  10. 2025Demystifying Long Chain-of-Thought Reasoning in LLMs5 Feb
  11. 2025s1: Simple test-time scaling31 Jan
  12. 2025Thoughts Are All Over the Place: On the Underthinking of o1-Like LLMs30 Jan
1 more
  1. 2024Do NOT Think That Much for 2+3=? On the Overthinking of o1-Like LLMs30 Dec

Other references · 27

  1. 2017Proximal Policy Optimization Algorithmscited by 14,441
  2. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  3. 2025Qwen3 Technical Reportcited by 8,195
  4. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 7,916
  5. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  6. 2024Qwen2.5 Technical Reportcited by 4,713
  7. 2023Let's Verify Step by Stepcited by 3,598
  8. 2023GPQA: A Graduate-Level Google-Proof Q&A Benchmarkcited by 2,748
  9. 2025DAPO: An Open-Source LLM Reinforcement Learning System at Scalecited by 2,678
  10. 2025Gemma 3 Technical Reportcited by 1,686
  11. 2022Fast Inference from Transformers via Speculative Decodingcited by 1,615
  12. 2025Understanding R1-Zero-Like Training: A Critical Perspectivecited by 1,276
15 more
  1. 2025Kimi k1.5: Scaling Reinforcement Learning with LLMscited by 1,193
  2. 2024DeepSeek-V2: A Strong, Economical, and Efficient Mixture-of-Experts Language Modelcited by 1,093
  3. 2025SimpleRL-Zoo: Investigating and Taming Zero Reinforcement Learning for Open Base Models in the Wildcited by 505
  4. 2025Towards Reasoning Era: A Survey of Long Chain-of-Thought for Reasoning Large Language Modelscited by 381
  5. 2025From System 1 to System 2: A Survey of Reasoning Large Language Modelscited by 319
  6. 2025Chain of Draft: Thinking Faster by Writing Lesscited by 290
  7. 2025LLMs Can Easily Learn to Reason from Demonstrations Structure, not content, is what matters!cited by 132
  8. 2023A Survey of Reasoning with Foundation Modelscited by 70
  9. 2024Unlocking the Capabilities of Thought: A Reasoning Boundary Framework to Quantify and Optimize Chain-of-Thoughtcited by 56
  10. 2025Unveiling the Key Factors for Distilling Chain-of-Thought Reasoningcited by 40
  11. 2025Towards Understanding Distilled Reasoning Models: A Representational Approachcited by 26
  12. 2025Thinking Machines: A Survey of LLM based Reasoning Strategiescited by 25
  13. 202570% Size, 100% Accuracy: Lossless LLM Compression for Efficient GPU Inference via Dynamic-Length Float (DFloat11)cited by 23
  14. 2025FAEDKV: Infinite-Window Fourier Transform for Unbiased KV Cache Compressioncited by 11
  15. 2025Quantize What Counts: More for Keys, Less for Valuescited by 9

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 9.

9 papers.

  1. 2026DART: Draft-Agreement Routing for Training-Free Adaptive Thinking Budgets in Hybrid Reasoning Modelscited by 7
  2. 2026Mid-Think: Training-Free Intermediate-Budget Reasoning via Token-Level Triggerscited by 5
  3. 2026Path-Lock Expert: Separating Reasoning Mode in Hybrid Thinking via Architecture-Level Separationcited by 3
  4. 2026A.X K1 Technical Reportcited by 0
  5. 2026HRBench: Benchmarking and Understanding Thinking-Mode Switch Strategies in Hybrid-Reasoning LLMscited by 0
  6. 2026Agent Team Work Zone: An Automated, Persistent Workspace for Long-Lived Claude Code Agent Teamscited by 0
  7. 2026Fusion Training for Mathematical Generalization in Large Language Modelscited by 0
  8. 2026Beyond Correctness: Benchmarking and Aligning Response Behaviors in Hybrid-Thinking MLLMscited by 0
  9. 2026Capable yet Parsimonious: Extracting and Characterizing Hidden Chain-of-Thought in Frontier Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2025Thinkless: LLM Learns When to Think
  2. 2025Rethinking Thinking Tokens: LLMs as Improvement Operators
  3. 2024Thinking LLMs: General Instruction Following with Thought Generation
  4. 2025Think Only When You Need with Large Hybrid-Reasoning Models
  5. 2025Asynchronous Reasoning: Training-Free Interactive Thinking LLMs
  6. 2025Reasoning Models Can Be Effective Without Thinking
  7. 2025OptimalThinkingBench: Evaluating Over and Underthinking in LLMs
  8. 2025Fast, Slow, and Tool-augmented Thinking for LLMs: A Review