Collections

A collection is a short, ordered list of models chosen by the platform for one job, for example image-landing-page. Apps show it as a quality slider: rank 0 is the fastest and cheapest rung, the last rung is the best. Each rung has a tier label such as Fast, Pro or Ultra.

GET /v1/ai/collections
GET /v1/ai/collections/{id}

{
  "id": "image-landing-page",
  "object": "ai.collection",
  "name": "Image landing pages",
  "description": null,
  "models": [
    { "rank": 0, "label": "Fast", "description": null, "model": { "id": "google/gemini-nano-banana-2.1", "...": "same shape as /v1/models" } },
    { "rank": 1, "label": "Ultra", "description": null, "model": { "id": "openai/gpt-image-2.5-sunburst" } }
  ]
}

• Rungs whose model is unpublished, expired or outside the project's model allowlist are hidden. A collection with no visible rung is omitted (and GET /v1/ai/collections/{id} returns 404 collection_not_found).
• Each rung embeds the full model entry, including pricing and request_constraints, so a client can show the price of every tier without another call.
• Collections are managed by the platform superadmin in the console (AI → Collections). Products read them by id; they never hard-code model ids.