GET /v1/models, the console’s model grid and Composer (nodus/auto) now read a catalog that is checked against
every upstream hourly: a model an upstream adds at a published price is listed under a lab/name id, with its
context window, output limit, capabilities and per-token rates, without waiting for a release.
A model its upstream stops serving is no longer listed or routed to; requests that name it get 503 until it
returns, and named endpoints on it report ModelAvailable false in the meantime.
When an upstream changes a price, the new rate applies to requests that start after the change. A request already
running is charged at the rate it was held at, and its receipt’s pricebookVersion names that rate.