Models

A model is a single LLM offered by a connected provider — gpt-4o, claude-sonnet-4-6, text-embedding-3-small, and so on. Each provider you connect exposes a list of the models it serves, and that list is what to11 routes against.

​

The provider model list

When you connect a provider in Project → AI Gateway → Providers, to11 discovers the models the upstream API offers. You choose which of those models the provider exposes to your project — its allow-list. Only models on a provider’s allow-list can be routed to. You can revisit a provider later, from the Providers list, to add or remove models from its allow-list as your usage changes.

​

How models are chosen in routing

A model name is the thing every routing decision keys on:

• In passthrough (L1), you call the gateway with a model name and to11 forwards the request to whichever connected provider serves that model.
• In a route (L2), you pick the model the route matches on, and the route’s targets each pair a model with a stored credential. The models you can pick are the ones your connected providers expose.
• In a function (L3), the named task resolves to a route or to a model; the model still has to be one a connected provider serves.

When you build a routing rule under Project → AI Gateway → Routing, the model picker is populated from the allow-lists of your connected providers. A model that no connected provider serves is unknown to the gateway and returns a 404.

​

Next steps