A stable high-volume inference workload needs predictable capacity. The selected model supports Provisioned Throughput. What should the team assess before purchasing it?

1. Remove all load testing because purchased capacity guarantees every application SLO.
2. Assume every model and customization type supports identical provisioned options.
3. Only the single largest burst, assuming unused capacity has no cost.
4. Measured demand, utilization, supported model capacity, and commitment economics.