Why smaller is often better
For a narrow task a small fine-tuned model is faster, cheaper and more predictable than a large general one. Latency improvements alone change what is feasible in a request path.
Typical applications
Routing support tickets, extracting fields from documents, tagging content, and moderating submissions — all high volume, all narrow.
The trade-off
They generalise poorly. Change the task and performance falls away sharply, which is exactly why evaluation matters before deployment.
