Photo: Tima Miroshnichenko / Pexels
A nearly four-hour outage knocked out 19 components across OpenAI's flagship products, adding to a growing list of service disruptions in 2026
OpenAI’s ChatGPT and Codex platforms went down for nearly four hours on September 3, taking conversations, image generation, code tools, and login functionality with them. The company confirmed the issue has been resolved, with all systems restored to full operational status.
The incident affected 19 separate components across both platforms.
What happened and how long it lasted #
The service degradation began around 10:58 AM ET and persisted until 2:56 PM ET. Core features hit by the outage included conversations, code tools, image generation, integrations, and even basic login functionality.
OpenAI posted an update to its public status page at 2:58 PM ET, stating it was “investigating the issue for the listed services.” That update came just two minutes after the disruption had already ended.
Following the resolution, OpenAI’s status page reflected uptime figures of 99.63% for ChatGPT and 100% for Codex over recent periods. For context, 99.63% uptime over a 30-day period works out to roughly 2.7 hours of downtime, which is almost exactly what users experienced on September 3 alone.
A pattern emerges in 2026 #
This wasn’t a one-off. Earlier on the same day, OpenAI had already flagged a separate issue involving “ChatGPT Work Mode High Error Rates,” meaning users potentially dealt with two distinct disruptions in a single day.
A previous notable incident occurred on June 11, when elevated GPT-5.5 errors cropped up within Codex. OpenAI publicly logs these incidents on its status page.
Why uptime matters more than ever #
Codex, OpenAI’s AI-powered coding assistant, integrates with the ChatGPT ecosystem to provide support via web, API, command line, and IDEs such as Visual Studio Code.
Enterprise customers evaluating AI providers typically look at service level agreements and historical uptime data. A 99.63% uptime figure for ChatGPT falls short of the “five nines” (99.999%) availability that mission-critical cloud services typically promise.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our