"Small language models are changing agentic AI. They're cheaper, faster, and purpose-built for specific tasks whilst keeping data secure and computational costs low."

This whitepaper argues that small language models (SLMs) are often sufficient, and in many agentic AI settings preferable, to large language models due to lower latency, reduced compute and cost, and greater deployment flexibility. The authors advocate SLM-first or heterogeneous agentic architectures (SLMs by default, LLMs invoked selectively), discuss barriers to SLM adoption, summarize supporting evidence and case studies, and provide a practical LLM-to-SLM conversion algorithm for migrating agentic applications.