Exam-style
A media company's Bedrock summarisation service runs on Claude Sonnet 4.6 and sustains around 400,000 input and 60,000 output tokens a minute, all day and every day. The product team has committed to a p99 latency target, throttled retries keep pushing the tail out, and the per-token bill keeps climbing. What should they move to?
Reveal the answer
E. Reserve input and output tokens per minute on the Reserved tier, sized to the sustained rate
Steady, high, round-the-clock volume with a latency commitment is what the Reserved tier is priced for. Input and output tokens per minute are reserved separately at a fixed price per 1K TPM, billed monthly on a one- or three-month duration, above a floor of 100,000 input and 10,000 output TPM that this workload clears. The reservation sits outside the on-demand quota and traffic beyond it overflows to the Standard tier rather than throttling, so the p99 commitment survives a spike. Claude Sonnet 4.6 lists Standard and Reserved on its model card, and neither Priority nor Flex. Provisioned Throughput is the older Model Unit reservation, billed hourly, and its supported base-model list stops several generations short of Sonnet 4.6, which leaves it serving custom models rather than current ones. Backoff manages throttling but moves neither the tail latency nor the bill. Savings Plans cover EC2, Lambda, Fargate and SageMaker AI, not Bedrock inference. Batch inference costs half as much but writes results to S3 asynchronously, so it cannot serve a p99 target.
Q. Steady high-volume Bedrock traffic with a latency commitment. Reserve capacity or stay on-demand?
A. Reserve it, on the Reserved tier. You reserve input and output tokens per minute separately, from a floor of 100,000 input and 10,000 output, at a fixed price per 1K TPM for one or three months. That capacity sits outside the on-demand quota, and traffic above it overflows to the Standard tier instead of throttling. Claude Sonnet 4.6 lists Standard and Reserved on its model card, and neither Priority nor Flex. Provisioned Throughput reserves Model unitThe billing block Provisioned Throughput is sold in – one unit delivers a fixed tokens-per-minute rate for a specific model. by the hour instead, and the base models it supports stop generations short of the current Claude line. That leaves it serving customised models. On-demand suits spiky or low volume, at a per-token price with no commitment.
Why? Match the pricing shape to the traffic shape. Steady and high favours a reservation, and which reservation you can take out is set per model on its card.