Codex not working? Check outage or setup

Check the live report first

Open the live Codex status report. Compare its last official incident and five component states with the time your failures began. A missing or stale component is marked unknown, not healthy. You can follow the linked incident to the OpenAI status overview for the original updates.

The overview reports aggregate availability. Its own note says that availability for an individual customer may vary by subscription, model and feature. A green page therefore does not prove that your account or client is working; it may simply lack a Codex-specific component report or a matching broad incident at that moment.

If the status page has no matching incident

Check the incident history for the time of your first failure. An incident might have been reported after you checked or might already be resolved. For example, OpenAI marked a September 25, 2026 Codex outage resolved after it affected Codex Web, API, CLI and the VS Code extension. That dated event is an example of what an incident record contains; it is not evidence that Codex is down today.

If neither page shows a relevant event, write down the exact message, failure time and time zone, client and version, selected model and whether a new short request fails. If you already have another Codex client on the same account, a single short comparison can show whether both surfaces fail. A difference narrows your report but does not by itself prove the underlying cause. Avoid sharing credentials or full auth files.

Decide what to do next

• A matching active incident is posted: save unfinished work and follow that incident’s updates. Do not assume a resolution time unless OpenAI gives one.
• The relevant incident is marked resolved, but you still fail: retry one small request and keep the timestamp and exact error for support if it persists.
• No matching incident is posted: follow the guide for the literal error. A disconnected stream and a model-capacity message have different diagnostic paths. A single error is not enough to declare a service-wide outage.

Do not clear configuration, repeatedly sign out or change billing solely because the service appears unavailable. Those actions do not tell you whether an outage exists, and this page cannot inspect your account.

Limits of this page

This page is a checklist for the exact problem you see. The live status widget shows stored official snapshots, but neither it nor the public overview can diagnose a specific account or estimate how many users are affected. A failure without a matching notice still needs individual troubleshooting. The September 2026 example is historical and resolved.

This guide is part of our Codex connection troubleshooting.

First check whether this is a Codex service-side problem → View live Codex status.