# How to price an endpoint when the buyer is an agent

Price an endpoint for agents by starting from your cost per call, checking what comparable endpoints charge, and pricing so the number is directly comparable. Agents compare on price and latency with no switching cost, so legibility matters as much as the level, and a ceiling-based price loses to a fixed one in automated comparison.

Published: 2026-10-14
Last reviewed: 2026-10-14
Publisher: Roundhouse AI Ltd.
Source: https://roundhouseai.io/resources/how-to-price-an-endpoint-for-agents

---

## What changes when the buyer is software

Every pricing instinct built around human buyers assumes friction that is no longer there.

| Human buyer | Agent buyer |
| --- | --- |
| Switching costs time and attention | Switching costs one config change |
| Brand and familiarity carry weight | Neither is represented |
| Compares a few options | Compares everything in the catalogue |
| Tolerates complex pricing | Skips what it cannot parse |
| Signs up once, stays by inertia | Re-evaluates every call if you let it |

The last row is the one that should shape your thinking. A human who subscribed in March is probably still subscribed in September because changing is a hassle. A program has no such attachment. If you are second-cheapest and the cheapest is adequate, you lose every call.

That sounds grim and it has an upside: it also means you can win a buyer instantly, with no sales cycle, by being the better option at the moment of comparison.

## Start from cost, not from competitors

Work out what a call costs you: compute, upstream APIs, data acquisition amortised over expected calls, and the facilitator fee. That is your floor and you should know it to a reasonable precision, because at these price points a mistake of a tenth of a cent is a large proportion of the margin.

Then look at what comparable endpoints charge. Not to match them, but to know whether you are in a different order of magnitude, which usually means you have misunderstood either the market or your own costs.

Reported median per-call prices across x402 endpoints in 2026 clustered in the low cents, with common price points for discrete tools like search or scraping running from roughly $0.001 to $0.05. Treat that as orientation rather than a target: your category matters more than the aggregate.

## Make the price legible

This is the part that is specific to agent buyers and the part most sellers get wrong.

**Prefer `exact` over `upto` unless cost is genuinely unknowable.** A fixed price can be compared directly. A ceiling cannot, and in automated comparison a ceiling of fifty thousand units looks more expensive than a fixed two thousand even if it typically settles at one thousand. [Exact against upto](https://roundhouseai.io/resources/x402-exact-vs-upto-schemes) covers when the ceiling is unavoidable.

**One price per endpoint, not a tier table.** Volume tiers assume an ongoing relationship the anonymous buyer does not have. Put tiers behind your keyed path if you want them.

**Publish the price before the call.** In your 402, in your catalogue entry, in your documentation. A buyer that has to pay to learn the price will pay a competitor who told it.

**Get the units right.** Amounts in your 402 are in the asset's smallest unit; catalogue entries are usually in dollars. Publishing these inconsistently makes you look either a million times cheaper or a million times more expensive, and neither gets you the sale.

> [!WARNING]
> The units error is the most common and the most damaging, because it does not fail loudly. An endpoint priced at `10` smallest units when it meant ten cents is selling at a hundred-thousandth of its intended price and will look extremely popular right up until someone checks the revenue.

## Should you price below cost to get started?

Rarely worth it, for a reason specific to this market.

With human buyers, loss-leading buys you a relationship that outlives the discount. With agent buyers there is no relationship to buy. The moment you raise the price to sustainable levels, every price-sensitive caller moves on, and you have paid for traffic that leaves.

What is worth doing is a **free tier with hard limits**, so a buyer can evaluate your output before paying. That costs you very little, converts buyers who would otherwise skip you, and does not train anyone to expect a price you cannot sustain.

## What about being paid in reputation?

Partly real, and more real here than in most markets, because the payment record is public.

Every settlement you receive is visible, so your endpoint accumulates a history of distinct buyers paying you over time. That history is what the next buyer reads when deciding whether you are worth trying, and it is the one credibility signal that cannot be manufactured cheaply at breadth.

This is a genuine argument for pricing to win volume early rather than to maximise margin per call, but it is bounded: what accrues is evidence of being paid, not evidence of being good. [Auditing a service's traffic](https://roundhouseai.io/resources/how-to-audit-an-x402-services-traffic) is what a careful buyer actually runs, and it rewards breadth of counterparties more than volume.

## Frequently asked questions

### Should I price per call or per unit of output?

Per call if the work is roughly constant, per unit if it varies enough that a flat price means overcharging most callers. Variable pricing needs the `upto` scheme and costs you comparability.

### How do I change a price without losing everyone?

You cannot do it gradually with anonymous buyers, since there is nobody to notify. Change it, and expect price-sensitive callers to re-evaluate immediately. Keeping a keyed path with announced pricing gives your committed buyers the stability the anonymous path cannot.

### Is there a minimum viable price?

Practically, yes: the facilitator fee and the gas underneath it. Below roughly a tenth of a cent on most chains you are working for the settlement layer. Deferred schemes that batch payments push that floor lower.

### Should my price match my subscription's per-call rate?

No, it should be above it. Matching gives committed buyers a reason to leave the plan, which cannibalises predictable revenue to win nothing. [Per-call against subscriptions](https://roundhouseai.io/resources/x402-vs-subscriptions-for-ai-products) covers the trade.

### How do I know if my price is wrong?

Watch conversion from 402 to settlement. A high rate of 402s that never come back is a price, a latency or a terms problem, and the three are distinguishable by which clients come back and which do not.

## Where to check this yourself

[/services](https://roundhouseai.io/services) shows the listed price distribution across indexed endpoints, which is the anchor worth having before you set yours. [/transactions](https://roundhouseai.io/transactions) shows settled amounts rather than advertised ones, and the gap between those two tells you what buyers are really paying as opposed to what sellers are really asking.

---

Roundhouse indexes x402 settlements and ERC-8004 identities into one public record.

- Every article: https://roundhouseai.io/resources.md
- Every surface on this site: https://roundhouseai.io/llms.txt
- The index, endpoints and schema: https://api.roundhouseai.io/v0
- A key of your own, $0.01 over x402, no account: https://api.roundhouseai.io/v0/test/x402
- Limits and the SQL surface: https://roundhouseai.io/docs/api