Codex model at capacity: what to check
A capacity message is not the same as an exhausted allowance or an unsupported model. Check the official incident history, then compare one small request across available models and clients before changing settings.
Start here
Check OpenAI's current status and record the exact error, time, model and client before retrying once.
What you see
The turn ends with Selected model is at capacity. Please try a different model. OpenAI has used this exact wording for a multi-model service incident in the past. That July 2026 incident is resolved; the same message today does not prove a new outage or tell you why your request failed.
This is separate from “usage limit reached” and from “model is not supported when using Codex with a ChatGPT account.” A user can have remaining allowance and still see a capacity error, as public reports document. Check the actual message before changing a plan, model setting or sign-in method.
Check the scope before changing settings
- Open the current OpenAI status page and check whether it lists a Codex incident. Use the incident history to check whether a past event overlaps the time of your failure. A green aggregate status does not rule out an individual problem.
- Record the time, exact message, selected model, client (desktop, CLI or IDE), client version and whether the failure occurs on a new short request. Do not share tokens or full authentication files.
- If another model is already available to your account, try one short request with it. If only one model fails, that narrows the observation. If all available models fail, repeated model switching is unlikely to add useful evidence.
- If you already use a second Codex client on the same account and network, compare the same short request there. A Windows report describes desktop failing while CLI succeeded. Other reports describe failures across clients. Neither pattern proves a universal cause.
Do not treat a visible percentage of remaining usage as proof of server availability. Conversely, the capacity message alone does not tell you whether that request counted against an allowance. The usage-limit guide covers an exhausted allowance; the model-access guide covers an unsupported selection.
What to do with the result
If OpenAI lists a current Codex incident, preserve your work and check its updates before repeating the request. If only one client or model fails, keep that comparison with your report. If every small request still fails and no relevant incident is listed, send the error, time, model, client version and comparison results through the product’s support or feedback path. An incident can appear on the status page after your first check.
Do not delete configuration, sign out repeatedly or buy more usage solely because of this wording. The sources do not establish any of those as a general fix. A separate IDE report confirms that the wording can appear there, but not that all clients share one cause.
Known limits of this guidance
The official incident cited here is historical and resolved. The September examples are individual reports, not a measured outage rate or a tested repair. This page cannot see your account or identify the underlying cause of a specific request; it gives you a safe way to collect the evidence that separates a service incident from a narrower failure.
This guide is part of our Codex sessions troubleshooting.
First check whether this is a Codex service-side problem → View live Codex status.
Useful Codex tools
Sources
Each source lists what it is used to support. Sources are re-read on the review schedule, not continuously.
Users are experiencing elevated errors when selecting models
The exact capacity error affected multiple models during a July 9 incident that OpenAI later marked resolved.
The official place to inspect recent reported Codex incidents; aggregate status cannot diagnose one account.
The official place to check current incident notices and Codex components; aggregate availability varies by customer.
Desktop capacity error with CLI comparison
One Windows reporter observed repeated capacity errors across models in desktop while CLI worked on the same account and network, with allowance remaining.
Capacity error across models without a usage-limit message
A separate reporter saw the literal error across tested models and clients without a usage-limit message; remaining allowance was not independently established.
An IDE user reported the same error across multiple models, without proving the same cause as desktop or CLI reports.
Related guides
How this page is checked
- Evidence level
- Reported - based on public reports, not reproduced here
- Last reviewed
- 2026-09-27
- Content updated
- 2026-10-03
- Symptoms indexed
-
- A Codex request fails with a selected-model capacity message
- Another model may show the same message even though usage remains
- A different Codex client may behave differently on the same account
Reviewer note
Separately reviewed all claims, the literal error, source dates and routing rule on 2026-09-27 after the draft pass. One report did not independently prove remaining allowance, so that claim was narrowed. No universal cause or fix is claimed.