Checked 5 minutes ago
2 separate problems in the last 48 hours, most recently 16 hours ago.
This longer record is here so the 48-hour picture above can mean something — without it we could tell you a problem happened, but not whether that is normal for OpenAI.
Something was impaired on 12.60% of our checks over that window. That is how often OpenAI was reporting a problem somewhere, which is not the same as how often it was unreachable. The longest outage we have on record ran 20h 42m.
The last 30 days of OpenAI, one row each.
| Day | Clear | Up | Checks | Started | Longest ended |
|---|---|---|---|---|---|
| 2026-09-17 so far today | 100% | 100% | 130 | none | — |
| 2026-09-16 | 94.77% | 98.26% | 287 | 1 | 51 min |
| 2026-09-15 | 83.28% | 99.3% | 287 | 2 | 11h 17m |
| 2026-09-14 | 66.55% | 100% | 287 | 1 | — |
| 2026-09-13 | 90.59% | 100% | 287 | 2 | 1h 30m |
| 2026-09-12 | 91.99% | 100% | 287 | none | 9h 38m |
| 2026-09-11 | 53.66% | 100% | 287 | 2 | 3h 32m |
| 2026-09-10 | 42.51% | 100% | 287 | 2 | 12h 9m |
| 2026-09-09 | 95.14% | 100% | 288 | 1 | 1h 12m |
| 2026-09-08 | 68.99% | 100% | 287 | 1 | 7h 28m |
| 2026-09-07 | 100% | 100% | 287 | none | — |
| 2026-09-06 | 100% | 100% | 287 | none | — |
| 2026-09-05 | 100% | 100% | 287 | none | — |
| 2026-09-04 | 83.97% | 100% | 287 | 1 | 3h 50m |
| 2026-09-03 | 91.29% | 99.65% | 287 | 2 | 2h 2m |
| 2026-09-02 | 100% | 100% | 287 | none | — |
| 2026-09-01 | 26.48% | 100% | 287 | 1 | 20h 42m |
| 2026-08-31 | 71.08% | 77.35% | 287 | 2 | 5h 25m |
| 2026-08-30 | 100% | 100% | 287 | none | — |
| 2026-08-29 | 100% | 100% | 287 | none | — |
| 2026-08-28 | 100% | 100% | 287 | none | — |
| 2026-08-27 | 82.58% | 100% | 287 | 1 | 4h 15m |
| 2026-08-26 | 100% | 100% | 287 | none | — |
| 2026-08-25 | 100% | 100% | 287 | none | — |
| 2026-08-24 | 100% | 100% | 287 | none | — |
| 2026-08-23 | 100% | 100% | 287 | none | — |
| 2026-08-22 | 100% | 100% | 287 | none | — |
| 2026-08-21 | 85.71% | 100% | 287 | 1 | 3h 31m |
| 2026-08-20 | 82.29% | 100% | 288 | 3 | 2h 59m |
| 2026-08-19 | 100% | 100% | 287 | none | — |
Clear is the share of checks that found nothing wrong at all; Up is the share that found OpenAI actually reachable — a vendor reporting one impaired component scores zero on the first and a hundred on the second, and both are true. Started counts outages that began that day and Longest ended the longest that finished on it, so a problem running across midnight is counted once, on the day it began, and measured on the day it stopped.
Rows recorded before 8 August 2026 do not carry a check count — the figure was kept and the denominator was not — so their percentages are shown without one rather than with a number borrowed from somewhere else. Rows before 9 August 2026 have no Up figure either: until then we recorded only whether something was wrong, not whether it was down, and the two are different questions.
| Started | Lasted | Severity | What happened |
|---|---|---|---|
| 2026-09-16 18:44 UTC | 51 min | having problems | Elevated errors in ChatGPT Work |
| 2026-09-15 23:50 UTC | 40 min | partly down | Elevated errors with gpt-image-2.5-flare |
| 2026-09-15 07:12 UTC | 44 min | having problems | Elevated errors from GPT-5.6 and GPT-5.6 Instant on paid plans |
| 2026-09-14 15:58 UTC | 11h 17m | having problems | Elevated errors affecting Work Mode in ChatGPT |
| 2026-09-13 16:56 UTC | 54 min | having problems | Elevated errors for ChatGPT users in Europe |
| 2026-09-13 09:26 UTC | 1h 30m | having problems | Codex GitHub Review and Pull Request Failures |
| 2026-09-11 16:22 UTC | 9h 38m | having problems | 1% of ChatGPT Work (mobile/web) turns are failing |
| 2026-09-11 07:53 UTC | 3h 32m | having problems | Elevated errors for ChatGPT users in Europe |
| 2026-09-10 19:49 UTC | 1h 41m | having problems | Elevated errors affecting ChatGPT Work |
| 2026-09-10 02:21 UTC | 12h 9m | having problems | Unable to open shared ChatGPT Project using direct link |
| 2026-09-09 15:13 UTC | 1h 12m | having problems | Increased Error Rate For Pro and Plus Plan Conversations |
| 2026-09-08 14:32 UTC | 7h 28m | having problems | Elevated errors for image generation |
| 2026-09-04 07:00 UTC | 3h 50m | having problems | Users in APAC region may face increased error in ChatGPT, Work, image generation, File upload, Voic, and Codex Cloud. |
| 2026-09-03 14:58 UTC | 2h 2m | having problems | Elevated errors across ChatGPT and Codex |
| 2026-09-03 00:04 UTC | 7 min | partly down | ChatGPT Work Mode High Error Rates |
| 2026-08-31 22:27 UTC | 20h 42m | having problems | Elevated latency in the Responses API |
| 2026-08-31 22:27 UTC | 15h 22m | having problems | Elevated latency in the Responses API |
| 2026-08-31 15:04 UTC | 5h 25m | partly down | ChatGPT Work seeing elevated errors and latency |
| 2026-08-27 18:00 UTC | 4h 15m | having problems | Users may experience an increase in error rates in ChatGPT Work on Web and Mobile |
| 2026-08-21 19:48 UTC | 3h 31m | having problems | Unexpected logouts for some ChatGPT web users |
Status pages report the vendor's view of their own service, and they lag real
outages — often by many minutes, sometimes by hours, because a human usually has to
confirm before anything is posted. They also can't see the two things most likely to
be wrong at your end.
The first is your own connection under load. A line that idles at 20 ms can
jump to 300 ms the moment something starts uploading, which drops you from sessions
and looks exactly like a server fault. Your plan speed says nothing about this; it's
called bufferbloat and almost nobody measures it.
The second is where everyone else in your group is. A service can be healthy in your
region and unwell in theirs, and if one person is hosting, their problem is everyone's
problem.
OpenAI being fine where you are does not mean it is fine for everyone in your group. The live dashboard resolves status against each person's own region, and will tell you which of your friends is in an affected area — plus who has the connection to host, which is usually a different question from who has the fastest plan.
Worth a look when more than one thing feels broken at once: if several are genuinely down together, the fault is usually something they share rather than anything at your end. We are not claiming these particular services fail together — the live dashboard says when they actually do.