Tokens-to-comprehension
Tokens-to-comprehension is the amount of documentation a model must read before it can correctly operate an API — the reading cost of understanding, denominated in the tokens an agent spends ingesting docs. Concise, well-structured, machine-readable documentation lowers it; sprawling or scattered documentation raises the price of every integration attempt an agent makes.
Agents pay for every token they read, and they read inside a finite context window. That makes documentation a metered resource in a way it never was for humans: an agent that must fetch and hold many pages before finding the auth scheme is spending real budget on navigation. Entry-point files like llms.txt exist precisely to route agents to the dense, relevant surface first.
Reading cost compounds with model tier into the full cost of comprehension. An API that demands extensive reading before a model can operate it is more expensive at every tier, and expense is a selection pressure — when an agent or its operator can choose, the API that is cheaper to understand has the edge.
Tokens-to-comprehension sits inside Discry's cost-of-comprehension framing. The piece Discry measures directly is the capability floor — the lowest model tier that reliably operates the API from its docs — and the discovery axis rewards exactly the surfaces that cut reading cost: llms.txt files that route straight to the reference, markdown mirrors, and parseable specs, all evaluated on a plain fetch, the way agents actually read.
What does an AI agent make of your API?
Find out in about a minute — no signup.
Discry your API — free