Why I tried to kill token billing (and why we kept it)
Source
stripe.com
Date
Context
Measuring an AI product's computing costs does not require making those costs the customer's pricing unit. A pricing essay published by Stripe argues for retaining tokenThe chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.Full definition →-level accounting internally while presenting charges through product operations, such as generated images or enriched datasets.
Its proposed bridge is a shared credit balance: different operations consume credits at different rates, while the company continues tracking model usage underneath. This lets the invoice describe work delivered and leaves model selection and margins inside the business. It is an argument about pricing design, rather than evidence that credits necessarily improve revenue or make bills cheaper.
Key takeaways · AI-distilled
The post argues per-token invoices expose your model mix and markups, defining your value as the gap over someone else's model costs. As models get cheaper and interchangeable, customers can challenge that markup or route around you.
Its suggested middle step is unified credits: keep token-level metering internally for routing and margin tuning, while customers burn down one credit balance per product operation, so the invoice shows work done rather than models used.
It separates three pricing metrics: inputs (costs such as tokens), outcomes (customer results like revenue) and outputs (objective, countable work such as a generated image), and argues outputs align AI pricing with value best today.
Outcome pricing, the author argues, needs monopoly-like control of the telemetry or very large contracts; otherwise attribution is disputed, such as whether an AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → or a sales rep closed a deal. Fin's per-resolution price is framed as an output.
Adoption can flip abruptly: hybrid seat-plus-usage pricing went largely unused for about 18 months, then by August 2026 roughly one in six Stripe users past key revenue milestones were using or rolling it out.
Terms in this piece · Glossary
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
