OpenAI's status page confirms a full Codex outage spanning Web, API, CLI, and VS Code extension on Sep 25, 2026 with API-key login as the workaround, exposing single-provider availability risk for coding-agent workflows until the incident resolves with timing and root-cause disclosure.

What is this?

Why it matters to Scott

Measured heat

How the heat travelled

Evidence (5) — ⭐ canonical anchor

[# OpenAI](https://status.openai.com/)

Subscribe to updates

Powered by

Availability metrics are reported at an aggregate level across all tiers, models and error types. Individual customer availability may vary depending on their subscription tier as well as the specific model and API features in use.

Issues with Codex

Identified·Full outage

Login via API key will unblock access at this time.

Fri, Sep 25, 2026, 11:19 PM

(6 minutes ago)

·

[View all updates](https://status.openai.com/incidents/01M3DCNWMW57HYK8FJ5FBFPA39#updates)

Affected components

10:58 PM

Now

Codex

4 affected components

Codex Web

Codex API

CLI

VS Code extension

Updates

Identified

Login via API key will unblock access at this time.

Fri, Sep 25, 2026, 11:19 PM

Identified

We have identified that users are experiencing elevated errors for the impacted services.

We are working on implementing a mitigation.

Fri, Sep 25, 2026, 11:03 PM(16 minutes earlier)

Identified

We have identified that users are experiencing a codex outage.  
We have identified the internal issue and are heading towards a mitigation.

Fri, Sep 25, 2026, 10:58 PM

Issues with Codex - OpenAI Status

Interpretation history

• 2026-09-29T16:30:57Z
  The episode's meaning has shifted from 'live sustained outage awaiting resolution' to 'closed incident with an uncaptured ending': engagement is at floor (0.33 pts/h, 122h old, 0 comments/h; the 71.5 peer percentile is an age-cohort coasting artifact, not live speed), and the ~7h average recovery pattern implies Codex recovered within hours of the last capture — yet resolution time, root cause, and workaround efficacy were never caught. The new Azure OpenAI EU item (2 pts / 0 comments, no date, page, or body) is too thin to broaden the claim into an OpenAI-wide serving-stack wave; it stays a lead, not a material fact, so the quiet clock keeps running.
• 2026-09-29T15:28:28Z
  evidence attached: hn.story.49893997 — Azure OpenAI outage in EU within the same window suggests the serving-stack incident wave extends beyond Codex itself, broadening the availability-risk claim if confirmed.
• 2026-09-26T03:27:01Z
  The incident has graduated from fresh report to a confirmed multi-hour full outage: independent user reports on HN and Reddit document Codex still down ~4.5h past identification, with no resolution or root-cause update captured yet. Meaning shifts from 'incident being confirmed' to 'sustained single-provider failure awaiting resolution disclosure' — a live-test of Scott's own stack that keeps it in the Anthropic-outage lineage; the velocity spike rides the initial burst (95.5/h peak, now ~16/h steady) on a status-page story whose comments were merged away, so heat stays medium rather than high.
• 2026-09-26T03:22:47Z
  evidence attached: reddit.post.1wqf70u — User report that Codex remained down late into the evening documents the incident's persistence for the open outage case.
• 2026-09-25T23:41:46Z
  origin walked (opencode/cheap-glm, conf 0.98): anchor hn.story.49851169 -> echo.other.72c8ad90d1 by OpenAI
• 2026-09-25T23:34:37Z
  grounded: converges/high — A live hit on his own stack, not just his topic: ask routes its default calls through codex via the LAN LiteLLM proxy and proposal runs resumable Codex jobs, so
• 2026-09-25T23:26:59Z
  case created — One bounded incident: the community-side report and OpenAI's own status-page confirmation (claim owner's artifact, so it anchors) are the same episode, currently in active mitigation.

Decision trace

• 10-05 13:42drop_targetsquiet through full ladder or over cap 8
• 10-05 09:02drop_targetsquiet through full ladder or over cap 8
• 09-30 02:30repriceThe episode's meaning has shifted from 'live sustained outage awaiting resolution' to 'closed incident with an uncaptured ending': engagement is at floor (0.33 pts/h, 122h old
• 09-30 01:28attachAzure OpenAI outage in EU within the same window suggests the serving-stack incident wave extends beyond Codex itself, broadening the availability-risk claim if confirmed.
• 09-30 01:28propose_attachAzure OpenAI outage in EU within the same window suggests the serving-stack incident wave extends beyond Codex itself, broadening the availability-risk claim if confirmed.
• 09-26 20:35review_screenjev screen: no material development (noul=0.04)
• 09-26 14:20sensor_dirtycomment_update
• 09-26 13:27repriceThe incident has graduated from fresh report to a confirmed multi-hour full outage: independent user reports on HN and Reddit document Codex still down ~4.5h past identification, with no resolution or
• 09-26 13:22attachUser report that Codex remained down late into the evening documents the incident's persistence for the open outage case.
• 09-26 13:22propose_attachUser report that Codex remained down late into the evening documents the incident's persistence for the open outage case.
• 09-26 10:21sensor_dirtyvelocity_spike
• 09-26 10:21sensor_dirtycomment_update
• 09-26 09:41promote_anchororigin walk conf 0.98
• 09-26 09:34groundA live hit on his own stack, not just his topic: ask routes its default calls through codex via the LAN LiteLLM proxy and proposal runs resumable Codex jobs, so a four-surface outage — with the report
• 09-26 09:26createOne bounded incident: the community-side report and OpenAI's own status-page confirmation (claim owner's artifact, so it anchors) are the same episode, currently in active mitigation.
• 09-26 09:24propose_attachOpenAI's own status page confirms the incident as 'Identified – Full Outage', the first-party confirmation for the case.