Do Dedicated Endpoints keep charging when idle?

No. Dedicated Endpoints are billed only while the GPU is active, metered per second, with no extra charge for startup time. With min_replicas = 0, the endpoint sleeps when idle and billing stops (default cooldown is about 5 minutes). Set min_replicas ≥ 1 if you need it always on — note that billing continues for as long as it stays active.

See our Dedicated Endpoints pricing doc for more information. https://friendli.ai/docs/guides/dedicated-endpoints/pricing#gpus-and-pricing