How FreeModel handles a provider failure

A gateway is useful when a provider fails mid-task. FreeModel keeps the request shape stable and moves the call to the next candidate instead of returning the provider error to your application.

What happens on a failure

The request arrives at one OpenAI- or Anthropic-compatible base URL with a gateway key. The gateway resolves the model name to a candidate list, putting free-eligible models first. If the first provider returns a quota error, a rate limit or a hard failure, the next candidate is tried.

What the caller sees

The caller gets one response and one bill. A failed provider call is not billed as a completed call. The console records usage per key and per day, so you can see which model answered.

What it does not do

Failover does not make a weak model stronger. If a task needs a specific capability, check the model list first. The gateway chooses among candidates; it does not add capability that a provider did not publish.

Try it without a key

The public model list at https://freemodel.online/v1/models can be read without a key. The free tier is 800 rounds per month and 30 per day, with no card required.