AIF-C01 Practice Question 55: Intelligent Prompt Routing

Intelligent prompt routing gives you *a single serverless endpoint to efficiently route requests between different foundation models within the same model family*. For each request Bedrock dynamically predicts the response quality of each model and forwards the request to the one offering the best combination of quality and cost; AWS lists "optimized response quality and cost" and "eliminates the need for complex orchestration logic" as the benefits. Simple messages land on the cheaper model, complex ones still reach the capable one, and the team writes no routing code (option B).

- A is wrong because Provisioned Throughput buys *dedicated capacity* on a model for a time-based commitment. It suits steady, high-volume traffic and guarantees throughput, but every request still runs on the expensive model; it does not make the cheap messages cheap. - C is wrong because batch inference is asynchronous: you submit a dataset and collect results later. It is a real discount, but a customer waiting in a support chat cannot wait for a batch job to finish. - D is wrong because prompt caching only reduces the cost of re-processing a *repeated prefix* such as a long system prompt. It helps, but every request is still answered by the largest model, so the underlying mismatch between message difficulty and model size is untouched.

Reference: Understanding intelligent prompt routing in Amazon Bedrock