Dedicated Model Hosting is live: reserved, single-tenant GPU capacity + a private serving plane so p95/p99, TTFT, throughput, and cost are governed by your provisioned hardware + tuned serving config.