A model must run once per night to predict next-day stock values. The input is 3 MB collected that day, and inference completes in under 1 minute. How should you deploy this model in Amazon SageMaker to meet these requirements?

A serverless inference endpoint is ideal for infrequent, bursty workloads like a nightly prediction job. It automatically scales compute resources, so you only pay for the inference duration, eliminating idle costs. Setting MaxConcurrency to 1 ensures that only one inference request is processed at a time, which is appropriate for a single nightly run. A multi-model serverless endpoint is for hosting multiple models on a single endpoint, which isn't specified here. Asynchronous inference is for large payloads or long processing times, but this model has a small input (3 MB) and fast inference (under 1 minute). A real-time endpoint with autoscaling to 0 still incurs some overhead for scaling up and down, making serverless a more cost-effective choice for truly infrequent use.