An ML engineer trains multiple models using the same ML framework and wants to host them in Amazon SageMaker while minimizing inference costs. Which hosting option is the most cost-effective?

A multi-model inference endpoint is the most cost-effective solution because it allows you to host multiple models behind a single endpoint and a single SageMaker instance. This significantly reduces infrastructure costs compared to hosting each model on its own instance. SageMaker dynamically loads and unloads models from memory as requests come in, optimizing resource utilization. Creating a multi-container inference endpoint is less efficient because each container still consumes resources, and it's typically used for scenarios where different models require different inference environments or pre/post-processing logic, not for cost-effectively hosting many similar models. Creating multiple single-model endpoints would incur the highest cost, as each model would require its own dedicated SageMaker instance.