PRODUCTION SCENARIO
A publisher fine-tuned a model in Amazon Bedrock to generate headlines. The custom model will serve a few hundred requests a day in unpredictable bursts. The company does not want to pay for idle dedicated capacity.
Which solution will meet these requirements?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: Create a custom model deployment and invoke it for on-demand inference.
Amazon Bedrock serves a custom model two ways: Provisioned Throughput, which reserves dedicated capacity billed by the hour, or a custom model deployment for on-demand inference, where you pay only for what you use and provision no compute. Bursty low volume favors the deployment. A six-month commitment bills for every idle hour. Custom Model Import is for weights customized elsewhere, so re-importing a model Bedrock already holds buys nothing. Batch inference returns results to S3 later, not as requests arrive.