Small Language Model (SLM)
A small language model (SLM) is a language model with far fewer parameters than the large, general-purpose language models – typically in the range of a few billion parameters. The providers of the large models generally do not publish their parameter counts, so the gap can only be estimated. SLMs are usually specialised in particular tasks or domains, need less computing power and can therefore run locally on laptops, smartphones or company servers instead of only through cloud APIs.
In practice
SLMs suit clearly bounded tasks such as classification, summarisation or customer service answers within a limited subject area, where a large model would be oversized. The advantages are lower cost per request, faster response times and more control over data, because the data does not have to leave your own systems. The drawback: for open, broad questions and complex world knowledge, SLMs are usually less capable than large foundation models and often need targeted fine-tuning on your own domain to deliver good results.