Models list now reports per-model endpoints and curated-voice flag
GET /v1/audio/models now returns endpoints and curated_voices per model, so a model picker can show only valid synthesis routes and reject unsuitable voice selections before the request reaches the server.
GET /v1/audio/models now tells you more about each model before you try to use it.
Each model in the response now lists its supported endpoints — the synthesis routes (/v1/audio/speech, /v1/audio/stream) it accepts. Pass it to an endpoint it doesn’t list and the server returns 400 before synthesising anything.
The curated_voices flag tells you whether only carefully-matched voices work with this model (true) or every catalogue voice works, including workspace clones (false). It is the difference between a curated voice roster and an open one.
Together the two fields let a model picker show only valid combinations and reject unsuitable selections before the request is sent.
See the docs changelog for the API reference.