Model fallbacks
Specify backup models when an upstream request cannot be served.
Choose backup models
model selects the primary model. The models array appends backup models in your chosen order. Provider fallback tries another endpoint for the same model; model fallback moves to a different model.
{
"model": "deepseek/deepseek-flash:cyberrouter",
"messages": [
{
"role": "user",
"content": "Summarize this maintenance note: the backup completed."
}
],
"max_completion_tokens": 256,
"models": [
"deepseek/deepseek-v4-pro:cyberrouter"
],
"route": "fallback"
}Each model must have an eligible connected provider. Compare the backup model’s price, context size, supported parameters, and enhancement profile before including it.
When fallback happens
Fallback is used for retryable failures such as upstream rate limits, provider errors, timeouts, and unavailable endpoints. Validation, authentication, and permission errors need a request or account change. They are not fixed by trying more models.
Once a passthrough stream has started delivering output, the gateway cannot replace it with a fresh model response. Handle the stream error and usage event in your client. Attempts that incurred provider usage can contribute to the request’s final fee.
The response’s provider and request receipt identify what served the call. cyberrouter.fallbacks counts preceding attempts; it is not a success-rate metric.