An ML engineer must deploy four models (each built with a different framework) into a SageMaker inference pipeline and allow clients to call invoke_endpoint for each model. Which cost-effective solution satisfies these requirements?

A SageMaker multi-container endpoint is the most cost-effective solution. It allows you to deploy multiple models, potentially built with different frameworks, behind a single endpoint. This reduces operational overhead and costs compared to managing multiple single-model endpoints. Clients can then invoke specific models within this endpoint. Creating a SageMaker multi-model endpoint is incorrect because it's designed for deploying many models (hundreds or thousands) that share the same inference container, not for a small, fixed number of models with different frameworks. Creating multiple SageMaker single-model endpoints would work but is less cost-effective due to increased infrastructure and management overhead. Running a SparkML job is irrelevant for deploying models to an inference endpoint.