Server errors across LLM providers

How server errors show up across Gemini, OpenAI, Anthropic, OpenRouter and more, and how to route around them.

5xx errors are the provider's fault, not yours: the request was valid but the backend serving the model was overloaded or down. They cluster during provider incidents, so a fixed retry loop against the same provider mostly amplifies the outage. Retrying the same request against a different provider is usually the fastest recovery.