We've identified a service that was starved of compute resources and addressed that problem. Service has been restored.

We're investigating issues with slow responses and HTTP errors from the Replicate API.