AI Error WikiSearch

Codex: Selected model is at capacity

Codex says the selected model cannot take the request.

Quick answer

“Selected model is at capacity” says the chosen model could not accept this request at that moment. It does not, by itself, say your personal usage is exhausted. In a Codex issue, users reported this notice while their accounts still showed remaining usage, sometimes across Desktop, Web, and CLI.

Exact message

Selected model is at capacity. Please try a different model.

What it means

Treat the message as a failed request until you inspect the run. One reporter tried changing models and starting fresh sessions without reliable relief. A commenter later counted hundreds of failed turns in one personal log; those are retries and failures for one account, not hundreds of affected users. If a different model works for you, that is a way to continue, not proof of the original cause.

What to try

  1. Save the prompt and inspect any files or actions from the failed run. Check the usage display, but keep its reading separate from the capacity notice.
  2. Wait briefly and try once more, or choose another available model if the task allows it. Avoid sending many identical long tasks without checking their side effects.
  3. If it continues, record the time, model, client, usage display, and whether a simple new session also fails. Give those details to support privately.

User report

The issue’s original reporter said the notice appeared intermittently across Codex Desktop, Web, and CLI despite remaining quota. Switching models and opening fresh sessions did not reliably clear it. Another person posted a large count of failed turns from a personal log. The public thread documents their experience; it does not establish a single cause for every account.

FAQ

Why do I see this with usage left?

The two displays answer different questions. Your usage screen reports allowance, while the error says the selected model could not handle the present request. The issue includes people who saw both at once. Check whether the notice repeats in a simple new session, and keep a screenshot of the usage display for support rather than assuming either screen is wrong.

Is this a rate limit?

The capacity wording does not say that you hit a personal rate limit. A separate quota or rate-limit message would be stronger evidence for that diagnosis. If your usage screen still shows allowance, report that alongside the exact capacity text. Do not infer a billing problem from this wording alone.

Will switching models fix it?

It may let you continue if another model is available. It did not reliably help the original reporter, who saw failures across several models and clients. Save the prompt and check for partial work before switching, especially if the task writes files. If the second model fails too, note both models and the time rather than cycling through them repeatedly.

Sources