Meridian Line's safety-review summarizer sees an unpredictable, spiky volume of requests — quiet for hours after an incident report is filed, then a burst of follow-up questions from the review board right after filing, then quiet again. Which Foundation Model APIs configuration best matches this traffic pattern while controlling cost?
Select an answer to reveal the explanation.
Short Explanation
Reserving capacity for a fire that only sometimes happens means paying for a fire truck to idle in the driveway all week. Pay-per-token is the option that only bills when the review board is actually asking questions, which is exactly this bursty, unpredictable pattern.
Full Explanation
Pay-per-token purchasing is the better fit for unpredictable, spiky, low-average-utilization traffic like this one, because there is no capacity to size or reserve in advance, and cost tracks actual token usage directly — during the long quiet stretches, cost is near zero, and during a burst, the service scales to meet demand without a pre-provisioned ceiling. Provisioned throughput sized for the peak burst and run continuously would mean paying for that peak capacity around the clock even during the many idle hours, which directly contradicts the goal of controlling cost for a spiky workload. Sizing provisioned throughput for 'the average' of quiet and busy periods is a flawed compromise — it would throttle or add latency during genuine bursts (since average capacity is insufficient for peak demand) while still paying for reserved capacity during quiet periods, giving the worst of both approaches. A fixed traffic split between two identical model versions does nothing to address capacity or cost at all; splitting traffic evenly between two copies of the same model does not create additional aggregate capacity in the way scaling does, and it is a version-rollout mechanism, not a load-management one. For the exam, the pattern to recognize is: spiky and unpredictable favors pay-per-token; steady and predictable favors provisioned throughput.