Codex CLI prompts API users to break their config (to use an unsupported model)
Resolved 💬 6 comments Opened Feb 11, 2026 by lattwood Closed Feb 13, 2026
💡 Likely answer: A maintainer (etraut-openai, contributor)
responded on this thread — see the highlighted reply below.
What version of Codex CLI is running?
0.0.0/main
What subscription do you have?
API
Which model were you using?
gpt-5.2-codex
What platform is your computer?
_No response_
What terminal emulator and version are you using (if applicable)?
_No response_
What issue are you seeing?
When starting codex cli after ya'll released gpt-5.3-codex, it prompts the user to change models!
Codex just got an upgrade. Introducing gpt-5.3-codex.
Codex is now powered by gpt-5.3-codex, our most capable agentic coding model yet. It's built for long-running, project-scale work, with mid-turn steering + frequent progress updates so you can collaborate while it runs (and it's faster too).
Learn more: https://openai.com/index/introducing-gpt-5-3-codex/
You can keep using gpt-5.2-codex if you prefer.
Choose how you'd like Codex to proceed.
› 1. Try new model
2. Use existing model
Use ↑/↓ to move, press enter to confirm
If you hit enter and then try to use Codex?
■ stream disconnected before completion: The model `gpt-5.3-codex` does not exist or you do not have access to it.
What steps can reproduce the bug?
Set it up for api key auth. Set the model to gpt-5.2-codex. Run codex and when it asks if you want to try gpt-5.3-codex, respond in the affirmative.
What is the expected behavior?
Don't ask the user to change the configuration to something that ultimately breaks the product, given the authentication mechanism in use.
Don't give the user false hope they don't have to wait a few weeks for it to be available via API.
Additional information
Looks like when gpt-5.3-codex was added, supported_in_api was false, but latest main has supported_in_api is now true, in models.json.
6 Comments
you I'm getting the following error
The modelgpt-5.3-codexdoes not exist or you do not have access to it.I'm using it through an AI gateay (portkey) but I dont think that is relevant, my org has been consistently billing > $5000 a month so I'm pretty sure I'm on a good tier
@narengogi, that's not related to this issue. Please use the
/feedbackslash command to upload your logs and open a new issue that includes the thread ID and supplies the other details requested in the bug report template.@lattwood, thanks for the bug report. This will be fixed in the next release.
Same here, uploaded the logs:
019c6806-3c0b-7192-9f70-ee92475ca10eIf I do
/modelthough, it doesn't showgpt-5.3-codexbut it seems trying to use it, since I accepted to try the new model.I switched back to
gpt-5.2-codexand that one works.Same here today, can use it
Any solution to this? Having the same issue with gpt5.3 codex