Every model’s rates are now its cost divided by 0.95, down from 0.875, including Indra’s routing call. Agent runs
on Nodus-managed models, machines, storage and egress are priced as before.
A model served from more than one source is listed once in GET /v1/models and the console, with the rates,
context window and capabilities of the cheapest one available now. If that one is unavailable, the listing moves
to the next cheapest, and back once it recovers.
A request is charged the rates it was accepted at, even when another source finishes it.
qwen/qwen3.8-27b-vision is now another name for qwen/qwen3.8-27b.