Sagemaker Endpoint save without running m5.large instance

If I delete an endpoint on a trained model, will I be required to re-train it? How can I save that model and endpoint. To be used later, without continuously running the m5.large instance.

Say, I want to run an inference 2 days later. How can I make sure to not be charged for the endpoint instance for the entire time and only be charged for resources used during inference?
I understand that Deploying an endpoint might require an m5.large instance, whereas using an inference might be done by t3.medium instance. Can you please share documentation on this? or codes/videos? I am trying to understand this concept.

The issue names no repository file, notebook, or test. Start by reviewing SageMaker endpoint deletion, trained-model persistence, and inference options; done would require a clearly scoped example or documentation covering retraining, endpoint costs, and delayed inference.

Written by the indexing model from the issue text.