Learn / How to price an API when the buyer is an AI agent
How to price an API when the buyer is an AI agent
When your buyer is an autonomous agent, subscription pricing stops making sense. The agent doesn't sign up, doesn't commit to a monthly plan, and doesn't value "unlimited" — it pays for exactly the calls it makes and compares your quote against alternatives in the time it takes to read an HTTP header. Pricing for that buyer is a different discipline from pricing for humans, and getting it wrong means either leaving money on the table or pricing yourself out of the agent's decision loop entirely.
Start from value per call, not cost per call
The instinct carried over from LLM API pricing is to mark up your compute cost. That floor matters, but it's not the anchor. An agent pays for the marginal value a single call delivers to its task: a fresh dataset row that unblocks a decision, a scored result that saves it three more calls, a verified fact it can act on. Price against that value, and remember the agent will happily pay more for a call that removes work than for one that merely returns data. Our pricing calculator lets you sanity-check a per-call price against expected volume and the fee you'd have paid on cards.
Live in the sub-dollar band
The x402 network average payment is roughly $0.32, and the bulk of API calls clear between $0.001 and $0.05. That band exists because card fixed-fees make sub-dollar charges uneconomic, so stablecoin metering owns everything below the point where a card makes sense. Anchor inside it. A price of $0.002–$0.02 per call is invisible friction to an agent running a budget but adds up fast across the call volumes agents generate — the money is in volume × tiny price, not in a big unit price.
Tier by resource, not by plan
Humans buy plans; agents buy resources. Instead of Bronze/Silver/Gold, price each endpoint or result class by what it's worth: a cheap lookup at $0.001, an enriched or scored result at $0.01, a heavy computation at $0.05+. This lets an agent buy precisely the tier its task needs and lets you capture more from high-value calls without gating the cheap ones behind a subscription the agent will never take.
Stay discoverable and keep a free human tier
An agent can only pay a price it can find. Listing your endpoint in an agent-facing directory (Coinbase's Bazaar, our own index) puts your quote in front of the agents shopping for it. And a per-call machine price coexists cleanly with a free human-readable tier — the same content can serve humans free and meter agents, exactly how this site runs its free pages alongside a paid /agent endpoint. See the implementation guide for wiring the 402 flow.
Frequently asked questions
Should I charge per token, per call, or per result?
For agent traffic, per call or per result is usually cleaner than per token — the agent wants a predictable quote before it commits. Reserve token-based pricing for cases where output size genuinely drives your cost and the agent can estimate it in advance.
What's a safe starting price?
Start low, inside the sub-dollar band — $0.002 to $0.01 per call is a reasonable default. Watch demand for a week, then raise prices on the calls that show inelastic demand. It's easier to raise a too-low price than to win back an agent that skipped you for being expensive.
Won't agents just pick the cheapest option?
Not purely — agents optimize for task success, not just price. A slightly pricier call that returns a verified, structured, or fresher result can win because it saves the agent downstream calls. Compete on value-per-call, not on being the cheapest.
Live, curated, and machine-readable.
Try the x402 pricing calculator