How well a model uses available RAM or GPU memory, allowing it to run on smaller or less expensive hardware.