Is Codex down?
How to check Codex's status properly
Codex (CLI and cloud agent) authenticates through ChatGPT accounts and calls OpenAI's API — incidents on either surface reach Codex.
OpenAI's status page tracks API, ChatGPT, and related surfaces as separate components; check which one an incident names.
Codex does get its own components: "Codex Web" and "Codex in ChatGPT Desktop" are listed separately, and the board above filters to those plus Responses and Login. There is no component for the Codex CLI, so CLI trouble surfaces through the API rows or not at all.
What Codex's incident history actually looks like
OpenAI's feed is dense and slow to clear. Twenty-five incidents in the 21 days to 13 August 2026, a median of 105 minutes to resolution, and a tail the other vendors here don't have: one in ten ran past eight and a half hours, and the longest stayed open for about 42.
Almost none of it is about Codex. Ten of the 25 were ChatGPT conversation errors, and only two named Codex at all — "Elevated Errors in Codex Review" and "Increase in errors on API, Codex and Work Mode". Most of what lands on this page belongs to somebody else's product.
That combination is what makes the page hard to read. A red banner on status.openai.com is usually image generation or ChatGPT chat, neither of which stops a Codex run. Filter to the Codex, Responses and Login rows and the signal gets a lot cleaner.
25 incidents, 23 July – 13 August 2026. Counted from status.openai.com/api/v2/incidents.json, pulled . The live board at the
top of this page is current; these numbers are a snapshot and will drift.
Is Codex down, or is it just you?
- A 429 during a listed incident is congestion, not your quota. Back off and retry before you go hunting through billing.
- Codex signs in through a ChatGPT account, so a ChatGPT-scoped incident can lock you out of a CLI whose API path is perfectly healthy. It is the most confusing failure mode on this tool.
- The reverse happens too: an API-key session keeps working while ChatGPT-auth sessions fail. If you have both configured, switching tells you which layer is down in about ten seconds.
- A red banner on the page is not automatically your problem. Check whether the incident names Codex, the API or Login — if it names Sora or image generation, keep working.
What to do while Codex is down
- API green but ChatGPT degraded? Codex CLI sessions authenticated via API key may keep working while ChatGPT-auth sessions fail.
- Rate-limit errors (429) during an incident are congestion, not your quota — retry with backoff before assuming you're capped.
- Cross-provider fallback (Claude Code, Aider) is the practical answer to a multi-hour OpenAI outage.
Frequently asked questions
How do I check if Codex is down?
Codex has no status page of its own — it lives on OpenAI's, which does list "Codex Web" and "Codex in ChatGPT Desktop" as their own components. The live board above reads status.openai.com's official API in your browser and filters to those two plus Responses and Login, so you skip the Sora and image-generation noise that fills the rest of that page.
How long do OpenAI incidents usually last?
Longer than the other vendors on this board. Across the 25 incidents in the 21 days to 13 August 2026 the median was 105 minutes, one in ten ran past eight and a half hours, and the longest stayed open for roughly 42. If a Codex-affecting incident is listed, plan around it rather than sitting and retrying.
Is Codex down when ChatGPT is down?
Often, yes — Codex sign-in and the cloud agent depend on ChatGPT infrastructure. A ChatGPT-scoped incident can block Codex even while the raw API stays green.