Small language model
Also called SLM
A compact language model built to run cheaply and fast, often on your own device.
Think of it like
Like a scooter instead of a truck — less hauling power, but nimble, cheap, and it fits in the garage.
Example
A phone keyboard runs a small on-device model to suggest words instantly without sending anything to a server.
How it actually works
Small language models trade raw capability for speed, cost, and privacy, and modern training and distillation have made them surprisingly strong on focused tasks. The bet is that most real jobs don’t need a frontier model — a well-tuned small one, running locally, is often good enough and far cheaper.
For product teams
Often the right call for narrow, high-volume, or privacy-sensitive tasks where a giant model is overkill.
For engineers
A parameter-efficient LM optimized for latency, cost, and edge deployment; frequently distilled from a larger teacher.
Related
- Language model — The general concept.
- Student model — How they’re often trained down.
- Frontier model — The big models they contrast with.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome