Small language models are compact language models designed to deliver useful performance with lower memory, latency, energy, and deployment requirements than frontier-scale models. This guide explains the mechanism, trade-offs, evaluation, and controls that matter in practice.