The Pricing of AI APIs

About this paper

I have been reading API pricing pages for sixteen years, and the conclusion that surprised me most is this: the most transparent class of API provider I found across thousands in the catalog was not the banks or the Fortune 500 — it was the AI labs. The companies selling the most hyped, most expensive APIs of the decade publish their prices more honestly than almost anyone. That inversion is the whole lesson.

This paper is two things at once: a market breakdown with real per-token rates across the flagship labs and the open-model hosts, the spread that shows you exactly where the commodity is, and the discount machinery (caching, batch, committed-use) the headline rate hides — and a pricing playbook for your own API, whether or not it touches AI. Because the lesson under the token economics is the one I have argued for sixteen years: pricing is not a revenue problem you solve once, it is a business-design problem you live with.

What's inside

1. The unit — why the token won, and per-token as direct monetization
2. The market — flagship rates, the open-model floor, and the spread
3. The levers — caching, batch, flex, committed-use: the real price is never the headline
4. The on-ramp — free tiers in a category that mostly skipped them
5. Where pricing hides — opacity, repricing trauma, and what transparency buys
6. Pricing your own API — plans before pricing, cost floor and value ceiling
7. Anti-patterns I watch for
8. Self-assessment
9. Where this is going — machine-readable pricing, FinOps, and pay-per-outcome

What you get for $25.00