I think a cheat code here is to route your requests to GCP's Vertex AI, which has stronger uptimes. Can use it as a fallback or main provider.
Caveats: 1) May not be economic for those on flat-rate Anthropic subscription plans 2) I work at Google.
Caveats: 1) May not be economic for those on flat-rate Anthropic subscription plans 2) I work at Google.
From developer experience, hosting them ourselves allows us to take advantage of our unique infra and deliver fastest time to first tokens of the providers.