IL-0121 OpenAI · 2025-07-09ai-infra
OpenAI deployed a faulty code change to a control plane cluster that mainly manages low-priority workloads. That cluster was also managing some production inference workloads, so the change degraded certain API model endpoints and reduced the availability of ChatGPT search. OpenAI rolled the change back to restore normal operation.
Afterwards OpenAI added safeguards that keep production inference workloads from being managed by that control plane cluster and strengthened testing for high-impact control plane changes. It is moving all existing production inference workloads under the production control plane cluster and building a circuit breaker meant to stop large-scale unintended changes.
Between 2:10 PM and 3:05 PM PDT on July 9, 2025, success rates of the gpt-4.1-mini model in the API dropped for about 20 minutes, and ChatGPT search returned many more errors than usual for about 35 minutes.
Symptom ? High error rate
Root cause ? Faulty deployment
–
–
–