An AI model with a small size in its neural network, the number of its parameters, and the volume of the training data. It requires less computational power and memory which makes it suitable for on-premises and on-device deployments.