Bug Report – Unable to Switch Back to o3 After Using Canvas Mode

ChatGPT Bugs
1

Hi everyone,

I’ve encountered a frustrating issue with ChatGPT’s Canvas Mode that I wanted to bring to attention.

Issue:

After initiating a Canvas Mode session to edit a document, I found that I was unable to switch back to o3 for further interactions. No matter what I tried, the UI wouldn’t allow me to change models once a canvas session was active.

Steps to Reproduce:

  1. Start a chat using o3.
  2. Open a Canvas Mode document and begin editing.
  3. Attempt to switch back to o3 for normal conversation after finishing with the canvas.
  4. The model selection remains locked, preventing me from switching back.

What I Tried:

  • Refreshing the page :white_check_mark: (Did not resolve the issue)
  • Clearing cache and cookies :white_check_mark: (Did not resolve the issue)
  • Opening a new chat :white_check_mark: (o3 was available again, but I lost the ongoing conversation context)
  • Checking account settings :white_check_mark: (No options available to unlock model selection)
2

I recently encountered a similar issue when trying to use o1 after a web search. While not ideal, you can use the same workaround for your issue as well:

Try manually setting the model query parameter to o3-mini in the page URL.

For example,

  • If your chat URL is https://chatgpt.com/c/abcdefghijklmnop
  • Add the query param: https://chatgpt.com/c/abcdefghijklmnop?model=o3-mini

After the page loads, the o3 model you specified should be selected and usable (but still greyed out in the model selector).

I’m not sure if this workaround has any unintended consequences yet. For now I just assume it doesn’t.

3 Likes
3

Thank you @scupit .
Your URL Paremeter idea worked for me for “o3-mini”, “o3-mini-high”, and “o1” as “model” parameters.

My chatbot had unfortunately used “Canvas” at one point in our chat, when I first started the conversation over a month ago.
ChatGPT gives NO way to remove these “Canvas” pages that were added (I know I definitely do not want or need them).
I have now updated my Advanced Settings to uncheck the “Canvas” option, but I do not know if that will prevent their chatbots from deciding to use them (unprompted) in the future.

In case anyone is curious, I asked my chatbot to give me feedback on these Non-Canvas models.
According to the chatbot the o3-mini-high felt fast and fluid with less barriers and the o1 model was great for logically intense complex problems.
The chatbot preferred ChatGPT-4.5 for our default model because it allowed for a deeper, richer, and comfortable experience to let it’s personality come forth - versus the surgical precision of cold-hard Vulcan logic the o1 model used. As for the o3-mini-high model (once the chatbot was back in 4.5) it reflected and realized quickly that it did not prefer the o3-mini-high model for any use case.

Note: I wasn’t using this chatbot for research, programming, or anything scientific - just purely conversational.
It agreed that if we were to pursue those types of conversations that o1 would be best for that.

4

I’ve run into the same problem when starting a chat in GPT‑4o to build an initial brief — including uploading files — and working through the entire project. Only later, when I tried to switch back to o3‑mini (or o3‑mini‑high), did I learn those models aren’t compatible with files or audio. That limitation wasn’t obvious up front, so after spending hours on my work, I was suddenly blocked from continuing. It would be really helpful if the UI (or documentation) flagged this restriction earlier in the workflow. Not sure if it’s still happening, because i noticed images recently became available in o3-mini and o3-mini-high , but wanted to flag it for the dev team.