A logistics company runs a nightly batch job that pushes roughly 50 million tokens through a Foundry-hosted model every night between 1 a.m. and 4 a.m., with almost no traffic the rest of the day. Operations needs a guaranteed, consistent level of throughput and latency during that fixed nightly window, and cost predictability matters more than minimizing per-request price. Which deployment option best fits this need?
Select an answer to reveal the explanation.
Short Explanation
Think about the difference between renting a car whenever you need one versus keeping your own car parked in your driveway. Pay-as-you-go is the rental: it is flexible and you only pay for what you use, but if everyone else wants a car that same night, you might not get one right away. This company doesn't have flexible timing, though, it has one predictable, heavy job every single night. What they actually want is guaranteed capacity waiting for them at 1 a.m., every night, without competing with anyone else's traffic. That's what reserving throughput ahead of time buys you: a known amount of horsepower set aside just for you, with a steady bill instead of a bill that swings with demand. Free evaluation tiers are built for kicking the tires on a small project, not for a production job moving 50 million tokens a night, and they'll hit a wall fast. And trying to squeeze a big predictable job through the smallest possible instance, queueing whatever doesn't fit, just recreates the same unpredictability problem you were trying to avoid in the first place.
Full Explanation
The correct answer is B. A provisioned throughput deployment reserves a fixed amount of model capacity ahead of time, which gives the operations team the guaranteed throughput and predictable latency they need during their fixed nightly window, along with predictable monthly cost rather than variable per-token billing. Option A is incorrect because pay-as-you-go scales with demand but makes no capacity guarantee; if other tenants place heavy load on the shared service at the same time, the nightly job could see degraded throughput and latency, which is exactly what the team wants to avoid. Option C is incorrect because free-trial or evaluation-tier deployments are meant for small-scale testing, not production workloads with tens of millions of tokens nightly, and typically carry usage caps that would block the job partway through. Option D is incorrect because queueing requests against an undersized SKU trades away the guaranteed timing the team needs; the batch job would simply run slower during contention instead of getting the reserved capacity provisioned throughput provides.