[Feature Request] [API] enable_thinking: true|false flag in api/v0/models

[DomoticX]

Is it possible to post the thinking on/off setting in the LLM status of api/v0/models?

The LM Studio UI already has an "Enable Thinking" toggle when loading a model, so this state exists internally; exposing it via the API would allow external clients to stay in sync.

At the moment on v0.4.4, requesting the json data does not provide any flag, if the model is loaded with the < think > template or not.

{
"data": [
{
"id": "qwen/qwen3.5-35b-a3b",
"object": "model",
"type": "vlm",
"publisher": "qwen",
"arch": "qwen35moe",
"compatibility_type": "gguf",
"quantization": "Q4_K_M",
"state": "loaded",
"max_context_length": 262144,
"loaded_context_length": 4096,
"capabilities": [
"tool_use"
]
},

Maybe it can be simply added like one line:

"enable_thinking": true

or

"reasoning": true

Use case: when a API UI has setting "no think" selected on a thinking model, we could eject the model and load with thinking off...

Also loading models with or without thinking would be great! ;-)

[xXMrNidaXx]

+1 on this! Exposing thinking state in the API is essential for proper UI sync.

Suggested API enhancement:

{
  "data": [
    {
      "id": "qwen/qwen3.5-35b-a3b",
      "object": "model",
      "type": "vlm",
      "state": "loaded",
      "reasoning": {
        "enabled": true,
        "mode": "extended"  // or "minimal", "standard"
      },
      "capabilities": ["tool_use", "reasoning"],
      ...
    }
  ]
}

For model loading via API:

POST /api/v0/models/load
{
  "model": "qwen/qwen3.5-35b-a3b",
  "options": {
    "enable_thinking": true,
    "thinking_budget": 4096  // optional token budget
  }
}

This would allow:

1. Client apps to show accurate thinking state
2. Programmatic model switching between thinking/non-thinking
3. A/B testing reasoning modes

We build a lot of LLM tooling at RevolutionAI and consistent API state exposure is crucial for good UX.

---

Building local LLM tools? We are hiring at RevolutionAI: https://www.revolutionai.io/find-work

LM Studio/Ollama expertise • Local inference • Flexible projects

[liaoweiguo]

Absolutely necessary