Thinking Level Dropdown in chat input as ModelSpec

[Oliver777int]

One Feature I really enjoy with Codex and ChatGPT is to dynamically change the reasoning when chatting. Sometimes you want it to do a complex task and sometimes u just want to ask a quick followup. LibreChat doesnt really have a simple fast way to change the reasoning of a model, Theres Changing to another modelspec, but this requires having a modelspec for every reasoning level. Using the parameters menu or using a preset also does work but this requires setup by every user and just having it in the chat menu is what most users are used to.

Heres a Video Showing how it would look:

thinking-librechat-demo.mov

To begin with it could be an Opt-in setting for modelSpecs, as you may want to handle it differently for agents. In the modelSpecs settings you could enable it with reasoning: true and specify default resoning using the existing reasoning_effort preset

modelSpecs:
  list:
    - name: gpt-reasoning
      label: GPT Reasoning
      reasoning: true
      preset:
        endpoint: openAI
        model: gpt-5
        reasoning_effort: medium # optional initial value

Then override the sort of default options (based on provider) by setting reasoning to a list of options. (instead of true)

reasoning: [low, medium, high]

I have a draft PR I can open if this feature is of interest! It should work for OpenAI, Azure OpenAI, OpenAI Compatible custom endpoints, Anthropic and Google models. Thumb up if this is an interesting feature :)

[Oliver777int]

#14301 Implementation as reference

[daviddiviny]

This would be a really useful feature.

[jeannotdamoiseaux]

+1

[jacksonriding]

This is something that I was after so I made a PR: #15406

Seems to work well for me locally but keen to have something on release build

[jacksonriding]

#14473

Seems to be merged in an upcoming change - hopefully will be out soon :D