OpenAI's September 29 outage ran five hours and 22 minutes across 30 components

OpenAI's September 29 outage ran five hours and 22 minutes across 30 components

OpenAI's status page logs an incident on September 29 that ran from 05:52 PM to 11:14 PM UTC, five hours and 22 minutes, and lists 30 affected components across ChatGPT, Codex and the API. OpenAI graded it "degraded performance" rather than an outright outage, with some users seeing "failed requests, difficulty logging in or signing up, and tasks that do not complete".

A ChatGPT prompt asking it to fix a dashboard loading bug, over a code diff

An OpenAI promotional image of a ChatGPT prompt reading 'Fix the dashboard loading bug and clean up the empty state', over a code diff. Image: OpenAI

No cause has been given. The closing update promises that "the detailed Root Cause Analysis (RCA) will be published in the next 5 business days", which from September 29 runs to October 6, so the explanation is due rather than late.

The count breaks down 12, 14 and four

Twelve of the 30 components sit on the API side, including Chat Completions, Responses, the Agents API and Realtime. Fourteen are ChatGPT surfaces, among them Conversations, ChatGPT Work, Deep Research and Connectors/Apps, the surface that ties plugins to your work and personal accounts, and the remaining four are Codex: Codex Web, the Codex API, the CLI and the VS Code extension.

Login appears twice in that list, once under the API group and once under ChatGPT, which matches OpenAI's own description of people having "difficulty logging in or signing up". The Compliance API and Sites, two surfaces most readers will never touch, are listed alongside Conversations and voice mode.

The first mitigation did not hold

OpenAI opened the incident at 05:52 PM UTC and widened it at 07:08 PM, adding the Agents API to the title. At 07:39 PM it posted "we have applied the mitigation and are monitoring the recovery", then posted the same sentence again at 09:09 PM, at 10:03 PM and at 10:47 PM.

Four identical monitoring notices over three hours and eight minutes is the shape of a fix that keeps coming undone. Resolution came at 11:14 PM, and the timeline is the only account of the evening OpenAI has published.

Support chat went down three minutes earlier

A separate incident opened at 05:49 PM UTC, three minutes before the main one, and OpenAI logged that "support chat remains unavailable" while pointing people to an email address. That incident ran its own course, with a mitigation notice at 09:50 PM and a resolution at 11:18 PM, four minutes after the main one, so anyone who tried to ask OpenAI what was happening while it was happening had to email instead.

The severity label stayed at degraded performance for the whole evening rather than moving to a full outage, so parts of each service kept answering for some people. OpenAI gives no figure for how many requests failed, and says its availability metrics are "reported at an aggregate level across all tiers, models and error types".

Until the root-cause report lands, the clock and the component list are the entire public record. OpenAI publishes no user numbers, no regional breakdown and no error rates for the incident, and it has not said what broke.

Leave a Comment

Your email address will not be published. Required fields are marked *