Model Fallbacks

Automatic failover when a provider is unavailable.

Automatic fallbacks

If the primary provider returns a 5xx error or times out, AllToken retries the request through the next available provider. Your application receives a successful response without any retry logic on your end.

Fallback chain

Each model has a prioritized list of providers. When the primary fails:

  1. Request is routed to priority-2 provider
  2. If that also fails, priority-3 is tried
  3. Continues until a provider responds or all are exhausted

View the fallback chain for any model on its detail page under "Available Providers".

Error passthrough

Client-side errors (4xx) are not retried — they indicate a problem with the request itself (invalid API key, malformed input, rate limit exceeded). Only server-side errors (5xx) trigger the fallback mechanism.