Model refresh is the process of updating a deployed model, often by retraining, recalibrating, or replacing it with a newer version. It is used to maintain performance, incorporate new data, respond to drift, or address defects and governance requirements.