Definition
Small language models (SLMs) are compact LLMs (often hundreds of millions to a few billion parameters) optimized for narrow tasks, local deployment, and cost-efficient inference versus frontier models.
Key Points
- 2026-07: cisco-antares 350M/1B security SLMs for vulnerability-localization — vendor claims competitive File F1 vs much larger models on vloc-bench
- Trade-off: specialization + locality vs general capability
- Fit for on-prem / air-gapped enterprise security workflows