OpenAI's services had a rough night. Starting at 1:35 p.m. ET on April 20th, the company began investigating a widespread outage affecting ChatGPT, Codex, and its API platform. Users reported widespread loading failures, which OpenAI officially classified as a partial outage. The status page listed 12 affected components for ChatGPT and one for Codex.
https://twitter.com/thsottiaux/status/2046367145588916687
The main disruption lasted about four hours and 13 minutes. OpenAI posted three updates on its progress at 3:48 p.m., 4:43 p.m., and 5:17 p.m. ET, finally announcing full restoration at 5:48 p.m. ET. During the same window, error rates for ChatGPT conversations spiked separately in Europe, starting around 3:50 p.m. ET and easing by 4:49 p.m. ET.
Earlier that day, starting at 6:13 a.m. ET on April 20th, ChatGPT Business also experienced issues for up to an hour when users tried to upgrade or add seats, with recovery being monitored from 8:43 a.m. ET.
A Brief Encore
Less than two hours after the main event was resolved, Codex went down again. Around 7:00 a.m. ET on April 21st, it suffered another brief outage.
Codex engineering lead Thomas Sottiaux posted on X that the incident lasted about 10 minutes. He noted that Codex itself helped during the fix, but the final, crucial insight still came from a human engineer. Sottiaux also said OpenAI would reset rate limits for affected users as compensation.