Small Language Model (SLM)
What is Small Language Model (SLM)?
A small language model (SLM) is a language model with far fewer parameters than a frontier LLM, usually trained or tuned for a narrow set of tasks, which makes it faster and cheaper to run and easier to govern.
Many contact-center tasks are narrow and repeated thousands of times a day: recognizing why a customer is contacting you, extracting an order number, summarizing a call, scoring sentiment or routing a message. A specialized small model can do this kind of work well, with lower latency and a much smaller compute bill than sending every request to a general-purpose model.
SLMs can also run in a private environment, which helps keep customer data where it belongs. The trade-off is breadth: an SLM is good at what it was built for and weaker outside it, so platforms typically orchestrate several models and send each task to the one that fits. Upstream Works’ UWXP uses specialized small language models alongside private LLMs.
Related: Large language model (LLM), AI models (SLMs and LLMs), UWXP.