Small language model (SLM)

Large language models run on expensive chips in data centers. A small language model has far fewer parameters, so it can run even on a laptop or a phone. Answers come quickly, and it takes less money and electricity to use.

It knows less and is weaker at complex problems, but it handles narrow jobs well enough, such as sorting documents by type, simple summaries and answering in a fixed format. So it has become common to split the work, giving easy jobs to a small model and hard jobs to a big one.

You do not need to carry every load in a big truck. For a local delivery, a small motorbike is faster and cheaper.

It lets you use AI on your device without sending your data outside. Interest has grown as companies trying to cut the cost of AI ask whether it really has to be a big model.