Between 19:05 and 19:18 UTC, some requests to the AI Gateway returned errors. We've since increased the capacity of the affected service, and error rates returned to normal levels at 19:18 UTC. We apologize for the disruption.