OpenAI suffered a full outage in the early morning for 4 hours, with ChatGPT, Codex, and the API platform all simultaneously unavailable.

According to Dynatrace monitoring, starting at 01:35 on April 21st, Beijing time, the OpenAI status page began investigating an incident that impacted the ChatGPT, Codex, and API platforms. Users reported widespread issues with loading, which the official investigation identified as a partial outage. The affected components identified include 12 instances of ChatGPT and 1 instance of Codex.

The outage lasted approximately 4 hours and 13 minutes. The official updates were provided at 03:48, 04:43, and 05:17 (Beijing time) indicating progress towards resolution, with full recovery announced at 05:48. During the same timeframe, the ChatGPT conversation error rate in the European region spiked, starting at 03:50 and mitigated by 04:49. Additionally, starting from 18:13 on April 20th, ChatGPT Business experienced a one-hour anomaly during seat upgrades or additions, with monitoring recovery beginning at 20:43.

Less than two hours after the initial outage was resolved, Codex encountered a brief incident around 7 am on April 21st. Codex's Engineering Lead, Thomas Sottiaux, later reported on X that the incident lasted approximately 10 minutes. He mentioned that while Codex itself contributed to the fix, the ultimate key insight came from human intervention. It was announced in parallel that affected users would receive a reset on rate limits as compensation.