Frontier model
What is Frontier model?
The most capable class of AI models available at a given moment, trained at the largest scale. New abilities tend to appear here first, and so do new risks, which is why frontier models draw the most safety testing and regulatory attention. Today's examples are the flagship models from the leading AI labs.
Why it matters
It is tempting to reach for the most capable model available, but the frontier is rarely the sensible default for a specific job. These models cost the most per call, almost always run on someone else’s servers, and change without much notice as the provider ships new versions. For a narrow, well-defined task, a smaller or older model is often cheaper, faster and more predictable, and it will not shift under you mid-project. The frontier is worth its price on genuinely hard, open-ended work that smaller models fail at. The skill is matching the model to the task, not paying frontier rates for something a modest model does well.
In practice
A team building an internal document summariser starts on the newest flagship model because it is the strongest on paper. The bill climbs fast, and testing shows a much cheaper model handles their straightforward summaries just as well. They keep the frontier model in reserve for the rare, genuinely complex request and route everything else to the smaller one. Capability was never the constraint here; cost per task was.
