GPT-5.4 mini immediately reports “Model at capacity” for all new and old Codex sessions

What version of the Codex App are you using (From “About Codex” dialog)?

Version 26.623.61825

What subscription do you have?

Business

What platform is your computer?

Darwin 25.4.0 arm64 arm

What issue are you seeing?

When creating a brand-new Codex session and selecting GPT-5.4 mini, the session immediately fails with:

“The selected model is at capacity. Please try a different model.”

This occurs before the repository is analyzed and before the prompt begins executing.

The issue is reproducible with completely fresh sessions, so it does not appear to be related to conversation length, context window usage, token limits, or repository size.

I intentionally use GPT-5.4 mini because it provides the best cost/performance tradeoff for long-running engineering work. When the model is unavailable, my only practical options are to wait an unknown amount of time or switch to a more expensive model.

What steps can reproduce the bug?

Feedback ID: 019f1379-be35-7530-9bc9-ed86736ecbf2

What is the expected behavior?

If GPT-5.4 mini is temporarily unavailable due to capacity, the application should provide a clearer indication of the service state, such as a queue or estimated wait time.

Ideally, users should be able to wait for the selected model rather than being forced to switch to a more expensive model or repeatedly retry.

Additional information

This issue has a direct workflow and cost impact.

I intentionally choose GPT-5.4 mini for large engineering workflows because of its lower inference cost. When it becomes unavailable, I either have to stop working or switch to a higher-cost model.

It would also be helpful if the application distinguished between:

• model capacity,
• temporary service incidents,
• and local/client issues,

so users can make informed decisions instead of repeatedly retrying or assuming something is wrong with their prompt.