Models and capabilities

The model catalogue is the authority for public IDs, lifecycle, operational availability, identity disclosure, capability evidence, limits, restrictions, and customer rates. Query the catalogue before presenting a model to a developer.

curl "$INFERENCEPASS_API_BASE/v1/models?limit=100" \
  -H "Authorization: Bearer $INFERENCEPASS_API_KEY"

GET /v1/models returns an OpenAI-shaped list. GET /v1/models/{model} returns the InferencePass detail, including capabilities, limits, restrictions, and verification_status. A model is not a first-party provider identity; read its disclosure before making product claims.

Project availability

An active project with an empty model allowlist can use published, operational models. A non-empty allowlist narrows that set. API-key model restrictions narrow it again. The request is rejected server-side if the selected model is outside either policy.

Capability states

Capabilities are supported, unsupported, or unknown; limits are known, unsupported, or unknown. supported and known require affirmative evidence. unknown is fail-closed for optional features.

API compatibility matrix
Surface	Status	Notes
`text_input / text_output`	supported	Required for a text request.
`streaming`	gated	The route must affirmatively support streaming.
`tools`	gated	The route must support function tools before tools are sent.
`structured_output`	gated	JSON mode and JSON Schema are admitted only when supported.
`image_input`	gated	Image parts are admitted only when the model has image evidence.

Availability, capability, and pricing data are snapshots. Cache with the response etag and refresh on a model_unavailable or unsupported_capability response.