[GH-ISSUE #12438] [model] Qwen3:4b Still "Thinks" When "think" in the Interface Is Set to False #54774

Closed
opened 2026-04-29 07:17:14 -05:00 by GiteaMirror · 5 comments
Owner

Originally created by @pinghe on GitHub (Sep 28, 2025).
Original GitHub issue: https://github.com/ollama/ollama/issues/12438

What is the issue?

Qwen3:4b still engages in "thinking" even when the "think" parameter in the interface is set to false.

Relevant log output


OS

No response

GPU

No response

CPU

No response

Ollama version

No response

Originally created by @pinghe on GitHub (Sep 28, 2025). Original GitHub issue: https://github.com/ollama/ollama/issues/12438 ### What is the issue? Qwen3:4b still engages in "thinking" even when the "think" parameter in the interface is set to false. ### Relevant log output ```shell ``` ### OS _No response_ ### GPU _No response_ ### CPU _No response_ ### Ollama version _No response_
GiteaMirror added the bug label 2026-04-29 07:17:14 -05:00
Author
Owner

@rick-github commented on GitHub (Sep 28, 2025):

Use the non-thinking version: https://ollama.com/library/qwen3:4b-instruct-2507-q4_K_M

<!-- gh-comment-id:3343285896 --> @rick-github commented on GitHub (Sep 28, 2025): Use the non-thinking version: https://ollama.com/library/qwen3:4b-instruct-2507-q4_K_M
Author
Owner

@somnifex commented on GitHub (Sep 29, 2025):

Check the API you are using; in my tests, the 'think' field is invalid when using the OpenAI compatible API, and you need to use the native Ollama API.

<!-- gh-comment-id:3344618362 --> @somnifex commented on GitHub (Sep 29, 2025): Check the API you are using; in my tests, the 'think' field is invalid when using the OpenAI compatible API, and you need to use the native Ollama API.
Author
Owner

@pinghe commented on GitHub (Sep 30, 2025):

The ollama-js client library and its ollama.chat interface are utilized. Both the qwen3:8b and qwen3:1.7b models can normally respond to the setting of the think parameter, while only the qwen3:4b model fails to do so.

<!-- gh-comment-id:3350552364 --> @pinghe commented on GitHub (Sep 30, 2025): The ollama-js client library and its ollama.chat interface are utilized. Both the qwen3:8b and qwen3:1.7b models can normally respond to the setting of the think parameter, while only the qwen3:4b model fails to do so.
Author
Owner

@somnifex commented on GitHub (Sep 30, 2025):

The ollama-js client library and its ollama.chat interface are utilized. Both the qwen3:8b and qwen3:1.7b models can normally respond to the setting of the think parameter, while only the qwen3:4b model fails to do so.

The latest Qwen3:4B model no longer uses mixed reasoning; instead, it offers separate instruct and think models. The default model is think. You need to download the instruct model, as described by the user above.

<!-- gh-comment-id:3350561744 --> @somnifex commented on GitHub (Sep 30, 2025): > The ollama-js client library and its ollama.chat interface are utilized. Both the qwen3:8b and qwen3:1.7b models can normally respond to the setting of the think parameter, while only the qwen3:4b model fails to do so. The latest Qwen3:4B model no longer uses mixed reasoning; instead, it offers separate instruct and think models. The default model is think. You need to download the instruct model, as described by the user above.
Author
Owner

@pinghe commented on GitHub (Sep 30, 2025):

thanks

<!-- gh-comment-id:3351695967 --> @pinghe commented on GitHub (Sep 30, 2025): thanks
Sign in to join this conversation.
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#54774