Skip to main content

Text Formatting Missing

Bug Reports

I using GPT 5 (normal and reasoning. and it seems that there is no text format, so that the text is hard to read. Here is the current text format.


But, sometimes the result give us the text format like this (which is what i want)


i think that this is a bug because it also happened when using o4-mini. can you guys fix this?

Status: Completed10 comments

Log in to comment and vote

Comments10

  • Amber Beach

    Here’s a screenshot of the behavior, both on GPT-5. Original chat on the left, forked chat after toggling off web search on the right.

    No matter how many times I reroll the left chat, formatting is always broken, but the one on the right always works

  • Mark Florkowski

    Team

    This is dependent on what the model “decides” to output. You can add a custom instruction in https://t3.chat/settings/customization to tell the models to use more formatting which may help.



    “Please format text with headings for better readability” or something similar

    • Red Spool

      Thank you for the reply. I have tried your solution. But apparently, its still not working as i want. I have attached the video of the proof. When i used GPT-5, they added the text format. while in the other hand, when using the reasoning model. its not use the text format.

      In the video, i’ve noticed that the formatted text is automatically changed when the output is generating. for context, im using safari (from the post) and arc browser (from the video)

      • Red Spool

        i’ve noticed that when i turned on the web search at the GPT-5, the text format is off. could it be bug?

        • Mark Florkowski

          Team

          The chat model does not have access to tool calls, so when you enable search it switches to the standard tool-calling capable version of the model to perform search. The models will have some differences in their output formatting based on how they are trained.

          Proof that the model can produce formatting, depending on the prompt:

  • Amber Beach

    I’m also encountering the same thing, having the web search toggle on seems to break the formatting, regardless of whether it used the web search or not.

    As the text is streaming in, I can see a brief flash of it being styled as it was meant to, but something about the response format breaks the markdown rendering

    • Mark Florkowski

      Team

      https://feedback.t3.chat/p/text-formatting-missing/comment/68a4c9c620bd4977e3ab7692

      Can you click the “Copy Response” button under the response that is missing formatting, and verify if the markdown contains any formatting characters (eg. # for headings, etc.)

      • Amber Beach

        Huh yeah you’re right, it really is just the model isn’t returning any markdown formatting.

        Very weird that the websearch toggle somehow makes the model use no headings.

        • Mark Florkowski

          Team

          The reason for this is it is a different model ( gpt-5 with minimal reasoning instead of gpt-5-chat) because the -chat variant does not have have tool calling support. The chat variant of the model is post trained to produce more “friendly” output which is why we use it.

          • Amber Beach

            Ah so that’s what they mean with the chat variant. Also explains why using gpt-5 in other tools like opencode results in a similar unformatted response.

            I unfortunately haven’t been able to make the T3 Chat customization prompt have it generate formatting, but requesting it directly in the message seems to work.

            Thanks for checking in on this!