Skip to content

How-ToGPU

How to Turn Thinking On or Off in Ollama (DeepSeek-R1, Qwen 3, gpt-oss)

Ollama lets you control reasoning models' thinking: --think and --think=false on the command line, /set think and /set nothink in chats, --hidethinking for scripts and a think field in the API. Check /api/show first, as models such as gpt-oss use low, medium and high levels instead.

By , Editorial team

2 min read

How to Turn Thinking On or Off in Ollama (DeepSeek-R1, Qwen 3, gpt-oss)

Reasoning models such as DeepSeek-R1 and Qwen 3 "think" before they answer, writing out a reasoning trace first. That can improve answers to tricky questions, but it also means waiting longer for a reply on a home PC. Ollama lets you switch thinking on or off, hide it, or (for some models) pick a level. This guide is based on Ollama's documentation and blog; Forge PC hasn't timed the difference on our own hardware.

Check what your model supports

Thinking controls vary from model to model. Ollama's docs say you can ask the /api/show endpoint, which returns a thinking object listing the supported values and the default. For gpt-oss, Ollama's example shows named levels, low, medium and high, with medium as the default. Other models use simple on/off (true/false) values, and values: [false] means the model doesn't support thinking at all. Ollama keeps a list of thinking-capable models on its website.

In the Ollama command line

Ollama's announcement of the feature lists these options:

  • Enable thinking: add --think to ollama run.

  • Disable thinking: add --think=false.

  • Hide the thinking but keep it on: add --hidethinking, for example ollama run deepseek-r1:8b --hidethinking "is 9.9 bigger or 9.11?". Ollama suggests this for scripting, when you only want the answer.

Inside an interactive chat

If you're already in an ollama run session, type:

  • /set think to turn thinking on

  • /set nothink to turn it off

Ollama's blog notes that thinking is enabled by default in the command line for DeepSeek-R1.

Through the API

Apps talk to Ollama through its local API, and both the chat (/api/chat) and generate (/api/generate) endpoints accept a think field. According to the docs:

  • true requests thinking output.

  • false requests no thinking. Models that can't switch thinking off use their lowest supported level instead.

  • null uses the model's default.

  • A string such as "low" or "high" picks a named level, if the model supports it. Use the exact value from /api/show.

The reasoning and the answer arrive in separate fields: message.thinking and message.content from the chat endpoint, or thinking and response from the generate endpoint. That separation is what lets front-ends collapse or hide the reasoning.

When to turn thinking off

  • Quick questions and chat: Ollama says that with thinking disabled the model will not think and directly outputs the content, which it describes as useful for getting answers fast.

  • Keep it on for maths, logic and code: Ollama says thinking can help a model work through different viewpoints to reach a more accurate answer.

For more ways to make local models feel faster, see how to speed up a local LLM in Ollama, and to measure the difference yourself, how to benchmark local LLM speed.

Our verdict

Use --think=false or /set nothink when you want quick answers, --hidethinking when you want the reasoning but not the clutter, and the API's think field in your own apps. Check /api/show first: models like gpt-oss use levels rather than a simple on/off switch.

The bottom line

Use --think=false or /set nothink when you want quick answers, --hidethinking when you want the reasoning but not the clutter, and the API's think field in your own apps. Check /api/show first: models like gpt-oss use levels rather than a simple on/off switch.

Sources

  1. 01Ollama Docs, Thinking
  2. 02Ollama Blog, Thinking

About the author

Forge PC Staff

Editorial team

The shared byline for articles researched, written and edited by the Forge PC editorial team. How we source, check and correct our work is set out in our editorial policy.

More from Forge PC Staff

Read next

How to Use Embedding Models in Ollama for Local Semantic Search and RAG

How-To

How to Use Embedding Models in Ollama for Local Semantic Search and RAG

Embedding models turn text into vectors for semantic search and RAG. Pull one such as embeddinggemma in Ollama, generate vectors with ollama run or the /api/embed endpoint, store them, then embed questions with the same model and compare using cosine similarity, all locally.

Forge PC Staff3 min

How to Turn On Variable Refresh Rate (VRR) in Windows 11

How-To

How to Turn On Variable Refresh Rate (VRR) in Windows 11

Windows 11's Variable refresh rate switch enables VRR for DirectX 11 full-screen games that don't support it natively, and Optimizations for windowed games helps windowed and borderless DX10/11 games use VRR. Turn on FreeSync or G-SYNC first; the VRR switch only appears with suitable drivers and a VRR monitor.

Forge PC Staff2 min