Elevated error rates - OpenAI Status

Write-up

Elevated error rates

Summary

On July 25th, 2026, between approximately 1:59 a.m. and 2:50 a.m. PDT, and again between approximately 4:24 a.m. and 4:58 a.m. PDT, some users experienced elevated errors while using ChatGPT, the API Platform, and Codex. An incompatible configuration update disrupted an internal service supporting authentication and other critical product functions. All affected services recovered by 4:58 a.m. PDT.

Impact

During this period:

Some users experienced elevated errors while accessing ChatGPT, the API Platform, or Codex.
Authentication, conversations, image generation, and other dependent product features were intermittently disrupted.
Two periods of customer impact occurred, with all affected services recovering by approximately 4:58 a.m. PDT.

Root Cause

A production configuration update introduced routing changes that were incompatible with the version of an internal service already in use. This prevented that service from retrieving essential user data, causing authentication and dependent requests to fail across ChatGPT, the API Platform, and Codex. After the initial rollback restored service, a separate undetected bug in the deployment system prevented the affected configuration from being properly blocked. As a result, an automated deployment reapplied the incompatible configuration, triggering a second disruption.

Resolution

Engineers restored service by rolling back the incompatible configuration, mitigating the initial disruption by approximately 2:50 a.m. PDT. When the configuration was automatically redeployed, engineers repeated the rollback and applied the correct deployment block. All affected services recovered by approximately 4:58 a.m. PDT. The incompatible configuration was removed, and the underlying deployment-blocking bug was subsequently fixed to prevent recurrence.

Prevention and Improvements

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

Prevent incompatible configuration updates. Strengthen validation so production configuration changes are checked against the service versions that must support them.
Improve early rollout detection. Extend testing and monitoring periods so availability issues are detected before a configuration change reaches all users.
Strengthen deployment safeguards. Ensure deployment blocks apply to the correct components and prevent blocked configurations from being automatically redeployed.
Expand pre-production checks. Improve staging validation and require successful health checks before critical configuration changes advance to production.

We apologize for the disruption and are continuing to strengthen the reliability of the systems that support ChatGPT, the API Platform, and Codex.

Availability metrics are reported at an aggregate level across all tiers, models and error types. Individual customer availability may vary depending on their subscription tier as well as the specific model and API features in use.