Small language model (SLM)

There is no fixed size threshold, but small language models are compact enough to run with modest hardware, sometimes entirely on-device. They trade breadth of knowledge and reasoning depth for speed, low cost, and the ability to run where data must stay local.

Small models often do well on narrow, well-defined jobs: classification, extraction, routing, short summaries, and simple tool calls, especially after fine-tuning on the specific task. They are much weaker at long, open-ended agent tasks that require planning across many steps.

In agent systems they often appear as helpers alongside a larger model, handling the high-volume easy steps while the larger model handles the hard ones. Evaluate them on your own tasks before relying on them, since performance varies widely.