Capability floor
The capability floor is the lowest model tier that can reliably operate an API from its documentation. It names the cost of comprehension: routing routine work to lower-cost models is standard industry practice, so an API that only frontier-tier models can understand is invisible where the volume is. Floors carry vendor-neutral labels — small-model ready, advanced-model required, or frontier required.
The floor exists because comprehension is relative to the reader. Documentation that a frontier-tier model operates cleanly can defeat a smaller model entirely, and production agent systems increasingly route routine work to the cheapest model that clears the task — model cascades and cost-based routing are standard practice. The floor tells you where, on that range, your documentation stops working.
For agent builders the floor is a pre-flight filter: it answers whether an API works at their model tier before an afternoon and a token budget get spent finding out. For API producers it is the commercial reading of docs quality — a low floor means agents can afford to understand you at scale.
Discry runs the same doc-grounded tasks across models at different capability tiers, and the floor is the lowest tier that reliably operates the API from its documentation. Profiles carry the vendor-neutral label — small-model ready, advanced-model required, or frontier required — and when coverage cannot support a floor, the profile discloses that rather than guessing. The exact models tested at each tier are disclosed in methodology and provenance.
What does an AI agent make of your API?
Find out in about a minute — no signup.
Discry your API — free