ChatGPT Outage History
Every incident OpenAI has published for ChatGPT recently, newest first, with how long each one lasted.
Showing 25 published incidents back to 2 Sep 2026.
- Minor17 Sep 202621 min
We are seeing elevated error rates across API models
- Major17 Sep 20261h 21m
Elevated errors affecting ChatGPT Work mode
- Minor16 Sep 202650 min
Elevated errors in ChatGPT Work
- Major15 Sep 202640 min
Elevated errors with gpt-image-2.5-flare
- Minor15 Sep 20261 min
Ads Manager login issues
- Minor15 Sep 202642 min
Elevated errors from GPT-5.6 and GPT-5.6 Instant on paid plans
- Informational14 Sep 202619 min
Degraded Performance affecting Agents API
- Minor14 Sep 202611h 13m
Elevated errors affecting Work Mode in ChatGPT
- Informational14 Sep 20261h 11m
Elevated error rates for Codex and ChatGPT Work
- Minor13 Sep 202651 min
Elevated errors for ChatGPT users in Europe
- Minor13 Sep 202628 min
Codex GitHub Review and Pull Request Failures
- Minor11 Sep 20269h 35m
1% of ChatGPT Work (mobile/web) turns are failing for existing threads
- Minor11 Sep 20262h 13m
Elevated errors for GPT-5.6 Sol on the API
- Minor11 Sep 20262h 23m
Elevated errors for ChatGPT users in Europe
- Informational10 Sep 20261h 55m
Delays in customer support responses
- Minor10 Sep 20261h 40m
Elevated errors affecting ChatGPT Work
- Minor10 Sep 202612h 6m
Unable to open shared ChatGPT Project using direct link
- Informational9 Sep 202624 min
Investigating unexpected usage limit resets
- Minor9 Sep 20261h 9m
Increased Error Rate For Pro and Plus Plan Conversations
- Minor8 Sep 20264h 37m
File uploads are delayed or failing
- Minor8 Sep 20267h 27m
Elevated errors for image generation
- Minor4 Sep 20263h 46m
Users in APAC region may face increased error in ChatGPT, Work, image generation, file upload, Voice, and Codex Cloud
- Minor3 Sep 20261h 57m
Elevated errors across ChatGPT and Codex
- Major3 Sep 20266 min
ChatGPT Work Mode High Error Rates
- Informational2 Sep 202631 min
Elevated errors creating new accounts
What this record shows
OpenAI reports openly and often. It posts short notices for problems many companies would never mention publicly, which is why this list is longer than you might expect for a service that is available well over 99% of the time. Most entries below are minor: a few minutes to a couple of hours, affecting some users rather than all.
That reporting habit is exactly why counting incidents is a poor measure of reliability. What matters is how much time was lost, not how many notices were posted — so read the durations rather than the row count. A month with ten brief minor incidents is a better month than one with a single four-hour outage.
One limitation to note: OpenAI does not tag its incidents by component, so this is OpenAI’s published record rather than a ChatGPT-only list. Some entries will have affected the API, Sora or another surface rather than ChatGPT itself — the titles usually say which. It is also why there is no separate Sora history page: the feed cannot distinguish one.
How to read the durations
Each entry runs from when OpenAI opened the incident to when it was marked resolved. That is a generous measure: an incident is usually opened once the problem is confirmed and closed after it has been stable for a while, so the window is wider than the period in which most people were actually affected.
Severity comes from the provider. Only major and critical entries count against the uptime figure on the ChatGPT status page, because a minor incident normally means degraded performance for some users rather than the service being unavailable.
This list is limited to what the provider has published and still exposes. Providers cap how far their public feed reaches, so an empty stretch further back means the record has aged out, not that nothing happened.
🔗 Keep checking