OpenAI · Error vs limit
Codex “Selected Model Is at Capacity”: Limit or Service?
This page covers the capacity-error class. Current five-hour and weekly allowances stay on Codex usage limits. After a real limit, choose wait, a banked reset, credits, or upgrade on reset vs credits.
Direct answer
Is this my Codex usage limit, or a service problem?
- What the message is
- Fact OpenAI’s status page has recorded Codex incidents titled
Selected Model is at Capacityas degraded performance, not as a usage-window exhaustion notice. - What it is not
- Derived The string is not, by itself, proof that a five-hour or weekly Codex window is empty. Included-allowance exhaustion is documented separately on the usage-limits guide. Check Settings → Usage or CLI
/statusbefore buying credits or upgrading. - What to do first
- Fact Official post-limit options (wait for the displayed clock, a banked reset if listed, credits on eligible Plus/Pro, or upgrade) apply only after a usage window is actually exhausted. If usage remains, switching models or retrying is the documented capacity recovery, not a purchase.
- This occurrence
- Unknown — account_dependent TraceCheck cannot see whether a specific current failure is regional capacity, model routing, context pressure, or something else. GitHub and forum reports are Observed demand, not a live outage board.
- Last verified
- Sep 10, 2026, 8:00 AM UTC
How to distinguish the two failures
| What you see | Class | Next check |
|---|---|---|
Selected model is at capacity. Please try a different model. | Fact Named on OpenAI status as a Codex degraded-performance incident class. | Read /status or Settings → Usage. If allowance remains, retry or switch models. Do not assume a purchase will clear it. |
| Usage window exhausted / reset time shown | Fact Plan allowance. Sizes stay on usage limits. | Wait, redeem a listed banked reset, buy credits if eligible, or upgrade. See reset vs credits. |
| Both at once, or generic “limit” | Unknown — insufficient context | TraceCheck does not guess. Use the account UI, not a community reconstruction. |
Official status vs a live outage
Fact OpenAI published a resolved Codex incident on 16 June 2026 whose title is the same capacity error. Later community threads and GitHub issues reuse the same string. Those later reports are Observed demand.
Unknown whether Codex is degraded for this account right now. This page is not a live blog. Check status.openai.com for current component state.
Observed Users report the message while Settings still show remaining usage. That pattern supports the capacity-vs-quota split; it does not prove a universal backend cause.
What does not fix it
- Buying credits or using a banked reset when
/statusstill shows remaining included usage. Those levers restore or extend allowance, not model serving capacity. - Treating a third-party “reset price” as an official capacity fix. Paid-reset SKUs remain Unknown on the reset guide.
- Assuming every model switch will succeed. Official wording is to try a different model; whether another model has capacity for this account is Unknown.
TraceCheck maintainer: review when OpenAI publishes a new Codex capacity incident, changes the error wording, or documents a different recovery path. Do not treat a named historical incident as a live outage.