ChatGPT and Codex are currently down for some users

OpenAI’s system status page confirms that some users are seeing elevated error rates when they try to use ChatGPT and Codex. The outage is partial rather than total: both products are affected, but the status language points to degraded reliability for a subset of traffic, not a full platform blackout. 9to5Mac reported the disruption after the status page made the elevated error rates public.

ChatGPT is the consumer and business chat product; Codex is the coding-oriented agent surface used for repository work, shell tasks, and iterative code changes. Elevated error rates typically show up as failed requests, stalled generations, or incomplete tool runs rather than a clean “service unavailable” page. Because both products sit on the same provider stack, a shared control plane, API gateway, inference pool, or auth path can raise error rates across chat and coding clients at once even when only some users hit the bad path.

For engineers and builders, that dual impact matters when ChatGPT and Codex sit in the critical path. Teams using Codex for refactors, test generation, or PR assistance lose a live coding loop when sessions fail mid-run. Product and support teams that depend on ChatGPT for drafting, triage, or internal tooling see the same class of request failures. Workflows that assume continuous availability need explicit retry, queue, or human-fallback behavior when the provider reports elevated errors.

In the broader market, OpenAI’s chat and coding tools are still default options for many developers and knowledge workers. When both degrade together, users often switch temporarily to other model APIs, IDE agents, or local tooling rather than wait on a single vendor. Partial outages also put pressure on OpenAI’s status communication: builders compare how quickly the status page updates, how specific the impact is, and whether ChatGPT and Codex are called out separately when error rates climb.

Practical takeaway: treat ChatGPT and Codex as best-effort dependencies during this window. Prefer idempotent jobs, save partial Codex work outside the session, and keep a second model or offline path for blocking tasks. Watch OpenAI’s system status page for a return to normal error rates on both ChatGPT and Codex before restoring automated or production-adjacent usage.