How well a model performs relative to its computational cost and resource requirements, important for deployment on devices with limited hardware.