The maintenance-triage agent's Model Serving endpoint gets heavy use during two shift-change windows each day and sits nearly idle the rest of the time. The team wants to avoid paying for reserved capacity during the long idle stretches while still supporting the busy windows automatically. Which endpoint setting fits this pattern best?
Select an answer to reveal the explanation.
Short Explanation
Don't rent a warehouse you only use twice a day — scale-to-zero is the setting that lets compute disappear when nobody's asking and reappear the moment the shift-change rush starts, which is exactly this workload's shape.
Full Explanation
Scale-to-zero is designed precisely for workloads with pronounced idle periods: the endpoint releases its compute when there is no traffic and automatically provisions again when a request arrives, trading a small cold-start latency hit for not paying for capacity that would otherwise sit unused most of the day. That matches the maintenance-triage pattern of two busy windows and long idle stretches. Provisioned throughput held constant all day is the opposite fit — it is designed for steady, predictable, latency-sensitive load and would mean paying for peak-sized reserved capacity around the clock, which is exactly the cost the team wants to avoid. A fixed traffic split between two model versions is a deployment/rollout control (used for canarying or staged promotion), not a cost-and-idle-capacity control, so it does not address this scenario at all. An AI Gateway rate limit caps the request rate an endpoint will accept for governance or cost-protection reasons; it does not release or reclaim idle compute, and setting it below the peak rate would actually throttle the legitimate shift-change traffic rather than solve the idle-cost problem. For the exam, map 'bursty with long idle gaps, cost-sensitive' to scale-to-zero, and 'steady, latency-critical, high volume' to provisioned throughput.