---
title: 'Small language model (SLM)'
description: "A language model small enough for a target deployment's memory, latency, cost, or control constraints relative to contemporary large models."
canonical_url: 'https://darkfactory.dev/glossary/small-language-model'
markdown_url: 'https://darkfactory.dev/glossary/small-language-model.md'
collection: glossary
date_published: '2026-08-04T00:00:00-04:00'
date_modified: '2026-08-26T00:00:00-04:00'
---

# Small language model (SLM)


## Definition

A small language model is a language model chosen or designed to fit a tighter resource envelope than contemporary large models. The envelope may concern memory, latency, energy, hardware, cost, or the ability to run inside a controlled environment. "Small" is relative to the time and deployment, so the label has no durable parameter threshold.

Parameter count affects storage and some compute costs, but it does not determine the whole footprint. Architecture, context length, numerical precision, cache size, serving batch, and active parameters all matter. Distillation and quantization can make a model cheaper to run without changing whether its original architecture was marketed as small.

SLMs are often trained or adapted for a narrower task set. A smaller model can run on an edge device, keep data within a local boundary, or meet a strict latency target. Those benefits depend on the complete system. Local execution can still require substantial memory, and a narrow model can still hallucinate, leak memorized data, or follow malicious instructions.

Model selection should start with a quality threshold on the actual task. Once several models pass, compare latency distributions, throughput, peak memory, energy or hosted cost, update cadence, and operational control. A small model that needs frequent fallback to a larger one may not reduce total system cost.

## Distinguish it from nearby terms

An LLM and an SLM are relative scale labels. A specialized model is defined by task scope and may be large or small. An edge model is defined by deployment location. The categories can overlap without meaning the same thing.

## Check your understanding

A 3-billion-parameter model handles 92 percent of requests locally, but the remaining requests fall back to a hosted frontier model. Which end-to-end measurements determine whether the SLM architecture lowers cost and exposure?

## Also called

SLM

## Related terms

- [Large language model (LLM)](https://darkfactory.dev/glossary/large-language-model)
- [Model distillation](https://darkfactory.dev/glossary/distillation)
- [Quantization](https://darkfactory.dev/glossary/quantization)
- [Inference](https://darkfactory.dev/glossary/inference)
- [Fine-tuning](https://darkfactory.dev/glossary/fine-tuning)

## Related factory areas

- [Model selection, routing & budgets](https://darkfactory.dev/factory/model-routing-budgets)

## Evidence and further reading

- [Small Language Models: Survey, Measurements, and Insights](https://arxiv.org/abs/2409.15790)