# How to price an x402 API for AI agents A practical pricing model for per-call agent APIs, with price bands, cost checks and the mistakes that make an agent skip a route. By APIStead Published: 2026-11-10 Updated: 2026-11-10 Category: Pricing Tags: x402, pricing, USDC, agent APIs, Base Pricing an API for AI agents is not the same as pricing a SaaS seat. The buyer is software. It makes a decision in milliseconds, often without a human in the loop. It compares the quoted amount with a budget, a task value and the cost of an alternative. That does not mean the price should be as low as possible. It means the price has to be legible. An agent can handle a higher price if the quote is clear, the unit is obvious and the expected output is worth the call. ## Start with the unit The unit is the first decision. Most agent APIs should price per successful call, not per token, per seat or per month. A per-call price has three advantages: - The agent can budget before it signs. - The buyer sees the cost of a workflow as a sum of calls. - The provider is paid for the work that actually happened. A subscription is a poor fit for an agent that may use an API ten times in a week and then not again for a month. An API key is a poor fit for an agent that has a wallet but no account. x402 removes both frictions: no key, no plan, no minimum. ## The three price bands The market has settled into a few loose bands. They are a starting point, not a law. | Band | Typical price | Examples | |---|---:|---| | Commodity lookup | $0.001-$0.005 | ZIP lookup, IP location, geocoding, weather, syntax checks | | Enrichment | $0.005-$0.02 | email verification, distance matrices, normalisation, scoring | | Evidence and compliance | $0.02-$0.10 | sanctions screening, pre-spend checks, audit receipts, document extraction | The middle of the public x402 market is around **$0.02 per call**. That is a useful anchor, but it is not a target. A commodity lookup at $0.02 will be skipped. A compliance check at $0.001 will attract low-quality traffic and may not cover the data cost. ## Price the outcome, not the compute A common mistake is to price the cost of the serverless function. The function may cost a fraction of a cent. The value to the agent is not the function. It is the decision the agent can make after the call. An email verification call can prevent a bad signup. A sanctions check can stop a payment. A pre-spend check can stop a wallet from signing a bad quote. Those outcomes are worth more than the CPU time. This does not justify a high price for a thin wrapper. The price has to match the evidence. If the API returns a score with no explanation, the buyer cannot tell whether it is worth anything. If it returns the matched list, the source timestamp and a receipt, the price is easier to defend. ## Add the costs that are easy to forget Before setting a price, list the real cost of one call: 1. Upstream data or API fee. 2. Facilitator fee, if any. 3. Base network gas, if the provider pays it. 4. Retry and failure rate. 5. Support and dispute time. 6. Verification and monitoring. 7. Refund or free-call policy. A route that costs $0.004 to run and sells for $0.005 is not a business at low volume. A route that costs $0.0005 and sells for $0.02 may be healthy, but only if the buyer keeps coming back. ## Make the quote readable The x402 challenge should answer four questions in one read: - How much is this call? - Which asset and network? - Who receives the payment? - How long is the quote valid? If the agent has to decode a custom field or call a second endpoint to learn the price, the route is harder to use. The challenge should be valid JSON with a stable schema and a payee that does not change between the quote and the settlement. ## Do not hide the price Some APIs return a 200 with a "price on request" page. That is a dead end for an agent. Some return a 402 with a price that changes after the agent selects a route. That is worse: the agent cannot budget, and the buyer cannot compare. Publish the price in four places: 1. The listing page. 2. The OpenAPI description or the route metadata. 3. The 402 challenge. 4. The receipt. If those four disagree, the agent will trust the challenge and the buyer will trust the receipt. The listing becomes the odd one out, and the support burden lands on the provider. ## Test the price with a real call A price is a hypothesis until someone pays it. Run one paid verification call per route. Record the amount, the network, the transaction hash and the response time. Then check the result against the listing. A useful test set: - A normal request that should succeed. - A malformed request that should fail before payment. - An unsupported network that should be rejected cleanly. - A request at the per-call cap. - A request over the cap, if the route enforces one. The goal is not to prove that the route can return a 402. The goal is to prove that a paying agent gets real product and a receipt. ## When to raise the price Raise the price when the route is reliable, the output is differentiated and the buyer keeps coming back. Do not raise it because a competitor charges more. Do not lower it to win a single high-volume agent that will leave when the subsidy ends. A better move is to improve the evidence: add a source timestamp, a confidence score, a matched list or a receipt. That makes the existing price easier to justify. ## When to lower the price Lower the price when the route is a commodity, the upstream cost is low and the main value is speed or convenience. A cheap route with high volume can be a good business if the margin is positive and the failure rate is low. Do not lower the price below the cost of a failed call. A failed call still consumes a request, a log line and sometimes an upstream fee. If the failure rate is high, the effective cost per success is higher than the sticker price. ## A simple starting point For a first-party API with a small upstream cost: - Commodity lookup: **$0.001-$0.003**. - Enrichment: **$0.005-$0.01**. - Compliance or evidence: **$0.02-$0.05**. Then measure. Keep the price stable for at least a month, publish the receipt and watch the repeat rate. A repeat buyer is a stronger signal than a one-off spike. The best agent API is not the cheapest one. It is the one an agent can understand, budget for and trust after the first call.