A small language model is a language model with far fewer parameters than the largest frontier models. It is quicker to run, cheaper per answer and can often run on a modest server or even a laptop or phone, while still handling many everyday text tasks well.
Smaller models are trained on large amounts of text, then often refined or distilled from bigger models to keep much of their capability in a smaller package. They generally have less broad knowledge and weaker complex reasoning, but they can be very good at narrow, clearly defined jobs.
Retailers and venues can use small models for classifying support emails, tagging products, extracting details from invoices, routing chat enquiries or powering simple assistants where speed and cost matter more than deep reasoning. Running one locally can also keep data inside your own systems.
The trade-off is capability. A small model may struggle with long documents, subtle instructions or unusual questions, and may hallucinate more on topics outside its training. Test it on your real tasks. Many setups combine both: a small model handles routine requests and passes difficult ones to a larger model.
Choosing the right size is one of the biggest levers on AI running costs and response speed, so bigger is not automatically better for a business.