PRODUCTION SCENARIO
A support assistant receives a mix of trivial questions and complex troubleshooting requests. The team sends every request to its most capable model. Analysis shows that 70% of the bill comes from questions a smaller model in the same family could answer well. The complex requests must keep their current answer quality.
Which solution will meet these requirements?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Use Amazon Bedrock Intelligent Prompt Routing for the assistant.
Intelligent prompt routing predicts the response quality of each model in the family per request and forwards it to the best combination of quality and cost, so trivial questions move to the smaller model while complex ones stay on the capable one. Provisioned Throughput bills hourly for the expensive model and still runs every request on it. Prompt caching cuts the cost of a repeated prefix but never changes which model answers. Model evaluation picks one model for all traffic.