A deliberately compact model, typically small enough to run on a laptop or a phone. It will lose to a frontier model on hard reasoning, and for a well-defined narrow job it is often indistinguishable, while costing a fraction as much, responding faster and keeping data on the device. A great deal of practical work does not need the largest available model.
Technical view
Usually in the range of one to a few tens of billions of parameters, produced through a combination of distillation, careful data curation and training on far more tokens than compute-optimal scaling would suggest. Quantised to four or eight bits they fit in consumer memory. The sensible pattern is routing: a small model handles the bulk of traffic and escalates to a larger one when confidence or task type warrants, which can cut cost substantially with little quality loss provided the routing itself is evaluated.
Related
Part of AI Terminology Index, section: Making Models Smaller, Faster and Cheaper.