Skip to main content

Not possible to select reasoning mode for GPT-5.4 in URL queries

Bug Reports

TLDR: When using /new url with params model=gpt-5.4-reasoning or model=gpt-5.4-instant the suffix is not respected and whatever mode you last selected in the UI gets used.

I use Kagi bangs to quickly open T3 chats.

I am hoping to have 2 bangs:

  • t3: 5.4 in instant mode

  • t3r: 5.4 in reasoning mode at medium settings

I have discovered that even though t3 chat unified 5.4 reasoning/instant under one model in UI, unlike 5.3. When creating chats through URL query parameters it still expect one of gpt-5.4-reasoning or gpt-5.4-instant.

Excerpt from error text when making a url with a bogus model name (I tried gpt-5.4 initially):
```json

{

"path": [

"model"

],

"message": "Expected \"gpt-5.4-instant\", actual \"gpt-5.4\""

},

{

"path": [

"model"

],

"message": "Expected \"gpt-5.4-reasoning\", actual \"gpt-5.4\""

},
```

However, in my testing the suffix is not respected and whatever I used last on t3 chat is what gets picked.

For example:

  • create a new chat, pick gpt-5.4 and set it to reasoning-medium

  • open a new tab with url
    https://t3.chat/new?model=gpt-5.4-instant&q=hello

  • Observe the settings are set to reasoning-medium

  • Expected: should be instant

Inverse:

  • In a chat with 5.4, pick instant

  • open a new tab with url
    https://t3.chat/new?model=gpt-5.4-reasoning&q=hello

  • Observe the settings are set to instant

  • Expected: should be some form of reasoning (I didn’t specify medium, so any level would be fine here, just not instant)

I have not yet tried to explicitly specify the reasoning level with an extra param since I don’t know if such a param exists, I will test that next, but I think the suffixes should be respected regardless of if one can control reasoning effort with a separate parameter?

Status: Completed4 comments

Log in to comment and vote

Comments4

  • Mark Florkowski

    Team

    There is an effort parameter:

    I’ll find a place to put this table in the app for reference

    I’ll also see about adding special cases for model slugs with -instant or -reasoning

    • Apricot Spoon

      There is a bug with the search_limit parameter normalisation, the limit always defaults to one, I have a post about it on my profile that is awaiting admin approval, please have a look through it Mark.

    • Maroon Quesadilla

      Apologies for the delayed response! Thank you, yes the reasoning parameters does exactly what I want. I think I would have been more likely to look for it if the model name didn’t include instant/reasoning.

      The following query strings work exactly like I wanted them to:

      Instant:
      https://t3.chat/new?model=gpt-5.4-instant&effort=instant&web=1&q=%s

      Reasoning:
      https://t3.chat/new?model=gpt-5.4-instant&effort=medium&web=1&q=%s

      Still confusing that I am using 5.4-instant for reasoning, but it works!

      I think my ideal resolution would be to replace gpt-5.4-instant/gpt-5.4-reasoning with simply gpt-5.4 since they both are equivalent (obviously not removing the old ones for backwards-compatibility purposes)

      • Mark Florkowski

        Team

        Yes, that is what I would like to do, just have not gotten around to cleaning up all of the places where those IDs are used, and making it back compatible