From January 1 to November 13, 2026, OpenAI's status data shows 99.93% uptime for ChatGPT conversations and 99.95% for ChatGPT login. OpenAI recorded 275 incidents with component impact in that period, of which 6 reached a full outage of a component. The longest major outage was July 23, at about 6 hours 40 minutes. OpenAI explains most of its outages, and the most common cause it gives is a configuration change.
Uptime
| Component | January 1 to November 13, 2026 | Past 90 days |
|---|---|---|
| ChatGPT conversations | 99.93% | 99.95% |
| ChatGPT login | 99.95% | 99.93% |
| ChatGPT Work | 99.90% | 99.90% |
| Agent | 99.85% | 99.55% |
| Image generation | 99.92% | 99.998% |
| ChatGPT group, all components | 99.63% | 99.26% |
| API group | 99.97% | 99.97% |
| Codex group | 99.97% | 99.94% |
The group figure is lower than the main components under it, so OpenAI's group number is not a simple average. The year-to-date column uses a custom date range on OpenAI's status data, not the default 90-day view.
Major Outages in 2026
Times are Pacific Time as OpenAI reported them, unless marked UTC.
| Date | What was affected | Time | Cause, paraphrased from OpenAI |
|---|---|---|---|
| Feb 3 | ChatGPT and the platform | 12:12 to 3:05 p.m. | A configuration change introduced an unexpected data type in a critical path |
| Apr 20 | ChatGPT, Codex and the API platform would not load | 7:18 to 9:00 a.m. | An overloaded internal authentication dependency |
| Jun 2 to 3 | Codex, ChatGPT and the Responses API | 9:46 p.m. to 12:01 a.m. | A configuration rollout made a service open connections repeatedly to a shared backend |
| Jul 14 | ChatGPT on the web | 4:39 to 5:19 p.m. | A routing change concentrated traffic on too little capacity |
| Jul 21 | Image generation | 4:52 to 7:47 a.m., with residual impact into the evening | A software update raised memory use and cut capacity |
| Jul 23 | ChatGPT, API, Codex and files, 19 components | 7:44 a.m. to 2:26 p.m. | A network maintenance failure at an infrastructure provider, then a second wave from that provider's automated DDoS protections |
| Jul 25 | 29 components | 1:59 to 2:50 a.m. and 4:24 to 4:58 a.m. | An incompatible configuration update hit an authentication service, and an automated deployment reapplied it |
| Aug 19 | chatgpt.com down, all sign-ups and logins | 4:50 to 5:19 p.m. | Two incompatible configurations applied to authentication systems |
| Aug 31 | ChatGPT Work | 7:40 a.m. to 12:53 p.m. | Demand exceeded available network capacity |
| Sep 3 | ChatGPT and Codex | From 7:43 a.m., mitigated about 8:20 a.m. | The routing system misread a configuration update and stopped sending requests to available services |
| Sep 25 | Codex, full outage | 3:33 to 4:47 p.m. | A false-positive security alert prompted an internal access change |
| Sep 29 | ChatGPT, Codex and API, 30 components | From 10:30 a.m., mostly mitigated by 11:00 a.m. | A feature rollout generated excessive internal network traffic |
| Oct 1 | Login, sign-up and ads | 11:39 a.m. to 12:28 p.m. | A shared storage system reached its capacity limit |
| Oct 15 | Mobile logins and sign-ups | 14:30 to 17:20 UTC | Not published |
| Nov 9 | API, 14 components | 23:08 to 23:50 UTC | Not published |
| Nov 11 | Logins, sign-ups and several features | 20:43 to 20:52 UTC | Not published |
Was ChatGPT Hit by the Microsoft Outage?
Only at the edge. On November 12, 2026 OpenAI opened a minor incident for its Microsoft 365 connector, saying new connections might break and tools might not work. No ChatGPT conversation or login incident is listed for that day. See the Microsoft 365 outage timeline.
Incidents Per Month
| Month | Incidents with component impact |
|---|---|
| January | 14 |
| February | 21 |
| March | 33 |
| April | 25 |
| May | 25 |
| June | 29 |
| July | 31 |
| August | 18 |
| September | 30 |
| October | 37 |
| November, to the 13th | 12 |
| Total | 275 |
By worst status reached, 6 incidents were a full outage of a component, 32 were a partial outage, and the rest were degraded performance.
What the Causes Have in Common
- Configuration changes and rollouts: named in seven of the 13 explained outages above.
- Authentication: the failing part on April 20, July 25 and August 19. When it fails, every product behind the login fails with it.
- Outside providers: the July 23 outage and a July 21 login incident both trace to an infrastructure provider's maintenance.
- Capacity: network or storage limits caused the August 31 and October 1 incidents.
OpenAI published 24 incident write-ups in 2026, more than any other assistant maker. For the comparison, see the AI assistant outage tracker.
How Many Users Report Problems?
Downdetector counts are user reports, and outlets often disagree. For the September 3 outage, one outlet reported more than 74,000 reports and another about 38,000. Use them as a sign that something happened, not as a measure of how many people were affected.
Methodology
Uptime, incident times and causes come from status.openai.com and its incident pages, read on November 13, 2026. Durations are from OpenAI's write-ups where one exists. Incident counts are our tally of incidents that recorded an impact on at least one component. Causes are paraphrased. Presenc AI did not measure uptime independently.