Summary

On July 21, 2026, some users experienced errors when generating images in ChatGPT. The most significant impact occurred between approximately 4:52 a.m. and 7:47 a.m. PDT, with residual errors continuing for a subset of requests until approximately 7:50 p.m. PDT. A second period of impact occurred on July 27, 2026, between approximately 6:20 a.m. and 1:22 p.m. PDT. An update to software supporting image generation increased memory usage and temporarily reduced available capacity. All affected functionality fully recovered by approximately 1:22 p.m. PDT on July 27.

Impact

During these periods:

• Some users were temporarily unable to generate images in ChatGPT.

• The API Platform was not affected.

Root Cause

An update to software supporting image generation increased memory usage and reduced available capacity. Requests shifted to the remaining systems, increasing load and causing additional failures. When additional capacity was brought online later, it encountered the same memory behavior, causing the second period of impact.

Resolution

Engineers first restored most capacity by reverting affected systems to a known-good software version, reconfiguring capacity, adding temporary capacity, and temporarily limiting excess requests while healthy systems came online. Error rates improved substantially by approximately 7:47 a.m. PDT, although residual errors continued for a subset of requests until approximately 7:50 p.m. PDT. During the second period on July 27, 2026, engineers added temporary capacity and deployed a corrected software version containing the memory fix. The deployment completed, and image generation fully recovered by approximately 1:22 p.m. PDT.

Prevention and Improvements

We have implemented or are working on several improvements to reduce the likelihood and impact of similar incidents:

• Catch GPU-memory issues before release. Add stress testing that simulates production traffic before software updates supporting image generation reach users.

• Strengthen deployment monitoring. Ensure monitoring and alerting are in place for every software update to production.

• Improve deployment management. Automatically pause or roll back any image that shows elevated error rates in production.

We apologize for the disruption and are continuing to strengthen the reliability of the systems that support ChatGPT image generation.