Just-in-time (JIT) Model loading via API ignores saved per-model settings

[tsebring]

LM Studio version: 0.4.1 (Build 1)

When a model is loaded by Just-in-time (JIT) on the local server (e.g. by request to /v1/chat/completions) it will NOT load the per model settings configured in "My Models" section in LM Studio. Both LM Studio and the online documentation states that it is supposed to load those per model settings.

See https://lmstudio.ai/docs/app/advanced/per-model

You can set default load settings for each model in LM Studio.

When the model is loaded anywhere in the app (including through lms load) these settings will be used.

See LM Studio My Models tooltips:

[strenkml]

I am also running into this, would be great if it was fixed

[rugvedS07]

Hello, thank you for the bug report! Can you please share following information, preferably screenshots:

1. What is the setting you have changed and what is the value you it being set as
2. The request you are using to JIT
3. List of all the models you have

[walczakp]

I'm having a similar issue:

1. I set Reasoning to Medium for openai/gpt-oss-20b
2. I call /v1/chat/completions which loads the model for me automatically. The request:

{
        "model": "openai/gpt-oss-20b",
        "messages": messages,
        "tools": tools
    }

3. I only have this model

Every time model is loaded with JIT, it loads with Low reasoning.
LM Studio 0.4.6 on Windows

[jarvis535claw]

Additional test:

I manually loaded the model in LM Studio, changed the API name to qwen3.6-27b, configured the custom load settings, and saved them as default.

Results:

• While the model remains loaded, calls to qwen3.6-27b correctly use the already-loaded model.
• If I unload/reload manually in LM Studio, the saved settings appear to persist.
• However, if I eject the model and then send a request via the OpenAI-compatible API so that LM Studio JIT-loads the model, it loads with default settings rather than the saved custom settings.

This suggests the issue is specifically with JIT loading not applying saved per-model default load settings, rather than only a LiteLLM/Hermes/model-name alias issue.

[techjanitor]

Yes this is still an outstanding issue. JIT loading from the OpenAI API loads the default model settings. Manually loading a model retains custom settings set in the models config. The most obvious limitations due to this are custom model jinja templates, batch size, etc