A fintech fine-tuned a supported Bedrock model on support tickets. Traffic is spiky, so the team wants to pay per use and avoid buying reserved capacity. Which inference path fits, where the model type and Region support it?
Show answer and explanation
Answer: A
Bedrock supports on-demand deployment for certain customized models in supported Regions. After the deployment is created, the application calls the custom model deployment ARN and pays for the usage it generates, which suits spiky traffic. Provisioned Throughput is the reserved-capacity option. SageMaker hosting and Custom Model Import both move the model onto different infrastructure and billing, so neither answers the requirement.
Why the other options are wrong
- B. Provisioned Throughput reserves model units and bills for them whether or not they are used, which conflicts with the pay-per-use goal.
- C. A SageMaker endpoint is billed by instance hours. That moves hosting out of Bedrock, which the team did not ask for.
- D. Custom Model Import brings externally trained weights into Bedrock. The team's model is already a Bedrock customization, so importing it again is unnecessary.