[Bug]: Missing reasoning toggle and API support for Qwen 3.5-4B models (GGUF)

[Milor123]

Which version of LM Studio?
LM Studio 0.4.6 (Build 1)

Which operating system?
Windows 11

What is the bug?
LM Studio fails to detect the Qwen/Qwen3.5-4B model (and its various GGUF quants) as a reasoning model. Consequently, the UI toggle to turn reasoning on/off does not appear.

Furthermore, when attempting to disable reasoning via the Local Server API, it throws an error. This issue seems strictly isolated to the 4B parameter version. The Qwen3.5-8B model is correctly identified as a reasoning model, displays the UI toggle, and successfully accepts API parameters to disable reasoning. I have tested different GGUF versions for the 4B model (including Unsloth's), but LM Studio does not recognize any of them as reasoning models.

Screenshots
Lmstudio model, dont thiking option found here.

Logs

2026-03-06 20:27:31 [DEBUG]
 Received request: POST to /api/v1/chat with body  {
  "model": "qwen3.5-4b",
  "system_prompt": "No pienses, se breve. Eres un traductor experto en... <Truncated in logs> ...traducción solicitada sin procesos de pensamiento.",
  "input": "Translate this New Adult hockey romance excerpt fr... <Truncated in logs> ... Maintain paragraph formatting. Text: Hello World!",
  "temperature": 0.3,
  "top_k": 20,
  "top_p": 0.9,
  "reasoning": "off"
}
2026-03-06 20:27:31 [ERROR]
 [LM STUDIO SERVER]  {
  "error": {
    "message": "Model 'unsloth/Qwen3.5-4B-GGUF/Qwen3.5-4B-Q4_K_S.gguf' does not support reasoning configuration.",
    "type": "invalid_request",
    "param": "reasoning",
    "code": "invalid_value"
  }
}

To Reproduce
Steps to reproduce the behavior:

1. Load the Qwen/Qwen3.5-4B GGUF model in LM Studio.
2. Look at the right-side configuration panel; notice the reasoning toggle is missing.
3. Start the Local Server.
4. Send an API request to the server attempting to disable reasoning.
5. See the API error in the server logs.
6. Swap to the Qwen3.5-8B model and observe that both the UI toggle and API parameters work as expected.

[will-lms]

Hello @Milor123 . You'll need to make sure you have the model.yaml artifact downloaded with the model, which is what our App uses to determine what configuration options are available.

You can do this easily in the GUI for our LM Studio Community quants, or by clicking "Use Model in LM Studio" on the qwen/qwen3.5-9b page.

If you want the unsloth quants specifically, you can follow their LM Studio Guide for the Qwen 3.5 model family.

Feel free to re-open if that does not resolve your issue.

[Milor123]

Hello @Milor123 . You'll need to make sure you have the model.yaml artifact downloaded with the model, which is what our App uses to determine what configuration options are available.

You can do this easily in the GUI for our LM Studio Community quants, or by clicking "Use Model in LM Studio" on the qwen/qwen3.5-9b page.

If you want the unsloth quants specifically, you can follow their LM Studio Guide for the Qwen 3.5 model family.

Feel free to re-open if that does not resolve your issue.

I think you didn't understand what I told you; first of all, I am referring to the Qwen3.5-4B model, the model that lm-studio uploaded, the qwen3.5-9b works perfectly ✅ (9B) not (4B ❌ )

The issue is with your lmstudio model Qwen3.5-4B ❌, the "thinking" toggle is not working @will-lms

I have the lmstudio version

[mayankbhagya]

I just deleted my old 4B model and redownloaded from lmstudio-community. Can confirm that this problem is real. Please fix.
I'm using LMS 0.4.7 on Ubuntu 24.04.

[Milor123]

Reopen the issue man! @will-lms

[acidSnkj]

Yes, it's strange, but the same thing happens to me too; I'm on Windows.
This is a bug and I believe there are other models affected

[mayankbhagya]

@will-lms checked the .lmstudio/hub folder. Missing all files such as yaml, json, readme...