A team wants to run multiple versions of a prediction model in production long-term, control the proportion of inferences served by each version, and minimize implementation effort. Which solution best meets these needs?

The correct answer leverages Amazon SageMaker's ability to host multiple production variants behind a single endpoint. This allows you to deploy different model versions (variants) and control the proportion of inference requests each variant receives by adjusting its weight. This directly addresses the need to run multiple versions long-term, control traffic distribution, and minimizes implementation effort by using a built-in SageMaker feature. Hosting each model with separate endpoints would require the application to manage routing logic, increasing complexity. SageMaker Neo is for optimizing models for edge devices, not for A/B testing or traffic splitting in a production endpoint. Batch transform is for offline processing of large datasets, not for real-time inference with traffic control.