Fourteen Incidents. Three Days. One Broken Desktop Update.
From Monday to Wednesday morning Pacific time, OpenAI's own status page logged fourteen incidents across ChatGPT, Codex and the API, and a desktop app release that couldn't start new Work threads.
A red traffic light, shown for illustration. Photo: Kevin Payravi, CC BY SA 4.0, via Wikimedia Commons.
You don't need a rumor mill to know when ChatGPT is having a rough week. OpenAI keeps its own receipt at status.openai.com, and between Monday and Wednesday morning it filled up fast: fourteen separate incidents. We went through every entry. All times below are Pacific.
Monday: conversations, Codex Cloud and API login
It started a little before 10 AM on Monday, October 5, with "Elevated errors in ChatGPT conversations." OpenAI said it had mitigated the issue just before 11 AM, but the incident wasn't marked fully recovered until about 5 PM. Codex Cloud had its own entry over the same stretch, where the status page said "users are experiencing elevated errors within Codex Cloud."
The evening didn't get quieter. From about 5:40 to 6:35 PM, developers hit "Elevated errors in API Platform login and administration APIs." Around 7:10 PM, an "Elevated Work Mode errors" entry closed out. Then, at 7:41 PM, OpenAI opened an incident with one of the longest titles we've seen on that page: "Elevated errors affecting multiple products including ChatGPT (conversations), Image generations, Spaces and Pages, API rehydration, and Dots." That one was over in about 36 minutes, which is quick, but it's still five products in one line. Two smaller entries landed Monday as well: "Issues saving and editing Pages" and an Admin Console that was "Intermittently Inaccessible."
Tuesday: Go users, PDFs and computer use
Tuesday morning it was ChatGPT Go's turn. From about 6:35 to 7:40 AM there were "Elevated error rates in ChatGPT Go conversations." Before lunch, the Responses API had "Errors processing large PDFs." From about 10:15 AM to 1 PM, OpenAI was "investigating issues for agents api and computer use."
Then, a little before 4 PM, workspace agents broke in a way that should worry anyone who leans on them for real work. OpenAI's own words: "users may not see the responses during the issue time." That incident ran about four hours before it was marked recovered around 8 PM.
The desktop update that couldn't start a thread
Tuesday evening, OpenAI posted "Unable to create new ChatGPT Work threads in latest version of desktop app," naming version 26.1002.51308. By about 11 PM it said it had found the problem and was "working on rolling back to the previous version." A fixed build shipped for Linux and macOS around 2:30 AM Wednesday. Windows users waited until about 4:50 AM, when OpenAI said affected users "should update to the latest version." That's close to ten hours of a paid desktop app that couldn't do the most basic thing it's for.
Wednesday: Codex again
As we write this on Wednesday, October 7, there are two fresh Codex entries. At about 10:21 AM OpenAI opened "Degraded performance in Codex and Work mode: Dot turn failures, delayed responses, and reduced proactivity," and a separate one warns "Dot in Codex may fail to create new threads." At the time we checked, OpenAI said it had applied a mitigation for the second and was still investigating the first.
And a September outage that only just closed
One more thing on that page caught our eye. A broad incident titled "Elevated errors across ChatGPT, Codex, and the API including the Agents API," which first went up on September 29 and described "failed requests, difficulty logging in or signing up, and tasks that do not complete," wasn't marked fully recovered until early Tuesday, October 6. OpenAI promised a root cause analysis within five business days. If you rely on these tools for paid work, that's the document worth watching for.
What to do when it happens to you
Check status.openai.com before you burn an hour debugging your own setup. If you're on the desktop app, update the moment OpenAI says a fix is out. And if a failed task ate your usage, screenshot the time and the error, because a refund or a reset is a lot easier to ask for with proof in hand.
Sources: OpenAI status history; ChatGPT conversations, October 5; Codex Cloud, October 5; API Platform login, October 5; multiple products, October 5; ChatGPT Go, October 6; computer use, October 6; workspace agents, October 6; desktop Work threads, October 6 and 7; Codex and Work mode, October 7; September 29 incident.