Skip to main content
The catalog lists open-weights models. Each model is served by one or more endpoints run by independent providers. You choose the model; TokenChange chooses the endpoint unless you pin one.

​
Browse the catalog

Anyone can browse models at tokenchange.io/models. Sign in and open Catalog to see each model’s endpoints and their rates. A model page lists its available endpoints with these columns: You can sort by speed and price and filter by quantization.

​
Availability

​
Capabilities

Endpoints list the features they support. Choose an endpoint with the capability your request needs.

​
Prices

Every endpoint sets three prices in USD per million tokens:
  • Input for prompt tokens.
  • Output for generated tokens.
  • Cached input for prompt tokens the provider served from its prompt cache.
See Credits and billing for how a request is priced.

​
Call a model

Pass the model’s slug as model. TokenChange routes the request to one of the model’s endpoints and moves it to another if that endpoint fails before responding.
On a model page, Use this model shows a ready-made request for that model. To list the models your key can call from code, use GET /v1/models on the inference API.

​
Pin a request to one endpoint

To send a request to one specific endpoint, append /ep/ and the endpoint’s identifier from the Endpoint column to the model slug:
On a model page, the Use action on an endpoint row shows a request already pinned to that endpoint.
Pinned requests never fail over. If the pinned endpoint is full, down or fails, the request fails instead of moving to another endpoint.
Pin when you need one exact deployment: a fixed price, a specific quantization, or consistent behavior for evaluation. Otherwise, let TokenChange route and use price options to stay within a budget.

​
Next steps

Routing and price options

Steer which endpoints serve your requests.

Inference keys

Restrict a key to the endpoints you trust.