A company used an Amazon Bedrock base model and trained a custom model to improve document summarization for internal use. What must the company do to use that custom model through Amazon Bedrock?

To use a custom model trained in Amazon Bedrock, you must purchase Provisioned Throughput. This allocates dedicated inference resources for your model, ensuring consistent performance and availability for your specific workload. Without Provisioned Throughput, your custom model cannot be invoked for inference. Deploying to an Amazon SageMaker endpoint is incorrect because custom models trained within Bedrock are designed to be used directly through Bedrock's API after purchasing Provisioned Throughput, not by deploying them to SageMaker endpoints. Registering with the Amazon SageMaker Model Registry is also incorrect; while SageMaker has a model registry, it's not the mechanism for making a Bedrock custom model available for inference in Bedrock. Granting access is a general security step but doesn't enable the model for inference; Provisioned Throughput is the specific requirement for operationalizing a custom Bedrock model.