Skip to main content
Models are selected in the request body. Your integration does not need a different route for each model.

​
List available models

cURL
The response is a list object:
The API may return more than one model as the catalog grows. Do not assume that the first item is the model you want; select by id and check status.

​
Read one model

cURL
Use the path parameter exactly as returned by the list endpoint. If the model does not exist, the API returns 404 with the standard error envelope.

​
Model status

Treat any status other than ready as a signal to stop sending new decision requests to that model and follow your fallback policy.

​
Public limits

The limits object describes model-level limits exposed by the API:
  • input_tokens — maximum input size in tokens.
  • candidate_tokens — maximum candidate-token budget.
  • choices — maximum number of choices accepted in one decision request or catalog.
The example values in the API schema are 4096, 128, and 300; read the live model descriptor rather than hard-coding them.

​
Revisions

The model field accepts identifiers such as goldenfish-g1 and revision-qualified strings such as goldenfish-g1@REVISION when a revision is available. Store the model string alongside each decision so you can reproduce which model produced a result.
Model IDs and limits are API data, not configuration constants. Discover them at startup or through a controlled configuration refresh and handle a model becoming unavailable between discovery and inference.