unable to use GPT-5.2

Resolved 💬 7 comments Opened Jan 27, 2026 by shayanultra Closed Jan 28, 2026

💡 Likely answer: A maintainer (etraut-openai, contributor) responded on this thread — see the highlighted reply below.

What version of Codex is running?

v0.92.0-alpha.3

What subscription do you have?

Pro

Which model were you using?

gpt-5.2

What platform is your computer?

macOS 26.3 Beta (25D5101c)

What terminal emulator and version are you using (if applicable)?

Terminal.app

What issue are you seeing?

Setting default Codex model to GPT-5.2 causes each new terminal session to prompt an offer to change model from GPT-5.2 to GPT-5.2-Codex. Despite stating no 99 million times in 99 million terminal sessions, Codex will ignore my decision and proceed to edit my config.toml before launching the terminal session:

[notice.model_migrations]
"gpt-5.2" = "gpt-5.2-codex"

Thus, it seems that I am unable to use GPT-5.2.

First of all, don’t prompt me an offer when /model works fine. Thus, the feature is completely useless, and a shame of why your time is spent on offering this zero-impact offering. And yes, it’s bugged - so a complete waste of your time. Adding features for the hell of it. Focus on shipping frontier technology and features instead of these fillers that cause headaches for everyone, foremost to you, the developer.

What steps can reproduce the bug?

Delete the feature altogether, go on a retreat, and come back ready to ship frontier technology or leave if you are not serious.

What is the expected behavior?

Clearly clicking on “no” should allow the usage of the current-selected model. But it gets forced into another model PLUS it injects and overwrites my config.toml. Seriously, what the fuck.

Additional information

Don’t ship junk.

View original on GitHub ↗

7 Comments

etraut-openai contributor · 5 months ago

We haven't heard of this behavior from other users. I wonder if there's something specific about your system or configuration that's causing this.

Try deleting (or temporarily renaming) your config.toml file and starting fresh with just model = "gpt-5.2" to see if that clears it up.

shayanultra · 5 months ago

I have a slight feeling this is intended on the developer’s end. If not, what is the setting to block the useless prompt that is trying to offer me a model change? Do you not think /model was sufficient? Changing the name of config.toml isn’t the solution here. I’ll wait until your team ships a setting to block the model change offer. And whoever shipped the feature, please, have them provide the motivation behind it. Thank you

etraut-openai contributor · 5 months ago

The intention behind this question is to notify users when a newer model is available. Most Codex users want to upgrade to newer models because they tend to be more capable, token efficient, etc. At the same time, we don't want to automatically change the default model because a user may wish to stick with an older model. That's why we emit the question one time and record the answer.

Are you seeing the question appear every time you start the CLI? If so, that's not expected behavior. When you launch the first CLI instance, it should write a [notice] entry to your config file noting that it has already asked you the question. Any CLI instance launched subsequently should refrain from re-asking the question.

You mentioned in the title that you're unable to use GPT-5.2. If that's the case, then that's unexpected behavior.

Could you clarify what behavior you're seeing?

guidedways contributor · 5 months ago

Stop removing this line - it doesn't do what you think it does. It's not setting your default model to 5.2-codex, it's an apparently poorly worded config essentially saying "we've informed the user about 5.2-codex". Do not edit or remove this line and you'll stop seeing the notice and should be able to use GPT-5.2 just fine.

[notice.model_migrations]
"gpt-5.2" = "gpt-5.2-codex"

shayanultra · 5 months ago

The intention behind this question is to notify users when a newer model is available. Most Codex users want to upgrade to newer models because they tend to be more capable, token efficient, etc. At the same time, we don't want to automatically change the default model because a user may wish to stick with an older model. That's why we emit the question one time and record the answer. Are you seeing the question appear every time you start the CLI? If so, that's not expected behavior. When you launch the first CLI instance, it should write a [notice] entry to your config file noting that it has already asked you the question. Any CLI instance launched subsequently should refrain from re-asking the question. You mentioned in the title that you're unable to use GPT-5.2. If that's the case, then that's unexpected behavior. Could you clarify what behavior you're seeing?

How often is a new model released? The rate of new model release by OpenAI is historically low such that when a new model does release, verberations are heard on a global scale. At least there should be a setting for notifications to be disabled, which would have only added no more than a minute to implement when deciding to ship this disguised-gimmick of a feature with a seemingly ulterior motive.

Stop removing this line - it doesn't do what you think it does. It's not setting your default model to 5.2-codex, it's an apparently poorly worded config essentially saying "we've informed the user about 5.2-codex". Do not edit or remove this line and you'll stop seeing the notice and should be able to use GPT-5.2 just fine. `` [notice.model_migrations] "gpt-5.2" = "gpt-5.2-codex" ``

How can we prove that "gpt-5.2" ≠ “gpt-5.2-codex"? If it’s verifiable, wonderful. Otherwise the config file strongly suggests that 'gpt-5.2' is an alias for 'gpt-5.2-codex': “migrations”, “=“.

shayanultra · 5 months ago

The intention behind this question is to notify users when a newer model is available. Most Codex users want to upgrade to newer models because they tend to be more capable, token efficient, etc. At the same time, we don't want to automatically change the default model because a user may wish to stick with an older model. That's why we emit the question one time and record the answer. Are you seeing the question appear every time you start the CLI? If so, that's not expected behavior. When you launch the first CLI instance, it should write a [notice] entry to your config file noting that it has already asked you the question. Any CLI instance launched subsequently should refrain from re-asking the question. You mentioned in the title that you're unable to use GPT-5.2. If that's the case, then that's unexpected behavior. Could you clarify what behavior you're seeing?

How often is a new model released? The rate of new model release has been historically low such that when a new model does release, verberations are heard on a global scale. At least there should be a setting for notifications to be disabled, which would have only added no more than a minute to implement when deciding to ship this disguised-gimmick of a feature with a seemingly ulterior motive.

Stop removing this line - it doesn't do what you think it does. It's not setting your default model to 5.2-codex, it's an apparently poorly worded config essentially saying "we've informed the user about 5.2-codex". Do not edit or remove this line and you'll stop seeing the notice and should be able to use GPT-5.2 just fine. `` [notice.model_migrations] "gpt-5.2" = "gpt-5.2-codex" ``

guidedways contributor · 5 months ago

Otherwise the config file strongly suggests that 'gpt-5.2' is an alias for 'gpt-5.2-codex': “migrations”, “=“.

Your understanding is incorrect. So are your prompts to ChatGPT. Instead, clone the repo, use codex and ask what would the following in my config file imply

[notice.model_migrations]
"gpt-5.2" = "gpt-5.2-codex"