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

Source: [Demystifying Hybrid Thinking: Can LLMs Truly Switch Between Think and No-Think?](<https://www.sota2.com/research/paper/68fa7c1c1e209131cea8c661>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2025-10-28T13:57:23.277Z

Authors: Shouren Wang, Wang Yang, Xianxuan Long, Qifan Wang, Vipin Chaudhary, Xiaotian Han

Published: 2025-10-14T16:19:44.000Z

## 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 \`\`\\texttt{wait}'' (from $5917$ to $522$ on MATH500). Our findings highlight the limitations of current hybrid thinking and offer directions for strengthening its controllability.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Mathematical Reasoning on MATH500 (Acc., Len., #Relf./Ans.)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-math500-acc-len-sharp-relf-ans.md>): Mathematical Reasoning; MATH500; Accuracy: 86.5.

- [Question Answering on GPQA Diamond (Accuracy, Length, Reflection Count)](<https://www.sota2.com/research/sota/question-answering-on-gpqa-diamond-accuracy-length-reflection-count.md>): Question Answering; GPQA Diamond; Accuracy: 41.16.

- [Mathematical Reasoning on AIME 24 (Accuracy, Length, #Relf./Ans.)](<https://www.sota2.com/research/sota/mathematical-reasoning-on-aime-24-accuracy-length-sharp-relf-ans.md>): Mathematical Reasoning; AIME 24; Accuracy: 30.

- [Question Answering on MMLU STEM (Acc, Len, Refl/Ans)](<https://www.sota2.com/research/sota/question-answering-on-mmlu-stem-acc-len-refl-ans.md>): Question Answering; MMLU STEM; Accuracy: 85.82.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2510.12680>)

[PDF](<https://arxiv.org/pdf/2510.12680.pdf>)