Am I charged when my deployment scales to zero?

No. With scale-to-zero enabled, a deployment at zero replicas incurs no replica charges. When the next request arrives, the autoscaler starts a replica and that first request waits through a cold start.

• A deployment sitting at one idle replica is not free — per-minute billing runs whenever replicas are up, traffic or not. If you're done testing, scale to zero or deactivate.

• Budget enforcement never stops dedicated deployments. A monthly budget can cut off Model API requests, but deployments keep running and billing regardless — see the Billing collection.

Docs: Autoscaling