OpenAI's AI coding tool Codex went completely offline for about 56 minutes on the morning of September 26, 2026 Japan time. The outage affected the web app, API, CLI, and VS Code extension, though ChatGPT itself was not impacted. After service was restored, OpenAI announced it would reset usage limits for all paid Codex and ChatGPT Work subscribers.

Timeline of the outage and recovery

Codex is OpenAI's AI coding agent, offered alongside ChatGPT, that can generate and review code and manage execution environments for developers. The outage was first confirmed at 7:58 a.m. Japan time, when OpenAI acknowledged the disruption and began working on a fix. By 8:19 a.m., the company said users could regain some access by logging in with an API key. After identifying the root cause and rolling out a fix, OpenAI confirmed full recovery at 8:54 a.m., putting the total outage at roughly 56 minutes.

Every Codex surface was affected, including the web app, API, CLI, and VS Code extension, leaving developers who rely on Codex in their workflows unable to code or run automated builds during the disruption. Other OpenAI services, including ChatGPT itself, were unaffected. OpenAI attributed the cause to an "internal issue" without disclosing further technical details.

Codex lead apologizes, resets usage limits

Tibo Sottiaux, who leads Codex at OpenAI, posted on X to acknowledge the incident, apologizing with the message "sorry about the brief disruption!" He announced that usage limits would be reset for all paid subscribers across both Codex and ChatGPT Work, effectively compensating users who lost tokens or had their work disrupted during the outage.

This is not the first time OpenAI has responded to a Codex outage this way. Sottiaux's X account shows a history of similar outage reports followed by usage-limit resets, suggesting that Codex's infrastructure has struggled to keep pace with its rapid growth in usage.

Some users report hitting limits faster after the reset

Despite the reset, some users on social media say they are now hitting usage limits faster than before. Several have reported that, since the post-outage infrastructure changes, they burn through their token allowance more quickly, and that certain models available through Codex, including Astra 6 and Sol 6, have become effectively unusable for some. That suggests the reset alone may not have fully resolved the underlying impact of the outage.

Summary

OpenAI's Codex experienced a roughly 56-minute, full-platform outage on the morning of September 26, but was restored the same day, and the company responded by resetting usage limits for all paid subscribers. With similar incidents recurring, balancing Codex's rapid growth with infrastructure reliability looks set to remain a challenge for OpenAI.

※Image is for illustrative purposes only.