What happens when you sell an API on x402? Lessons from the Bazaar and one small seller
Most x402 sellers earn almost nothing: in Coinbase's Bazaar catalog on 27 August 2026 the median price was one cent, one gift-card seller took about $21,000 of the catalog's $30,600 monthly revenue, the top ten resources took 78%, and the other 14,771 active resources shared about $6,800, under fifty cents each. Adding a price to a web service with x402 takes an afternoon. Selling something a machine will keep buying is the hard part. This article is the Group's account of what sold, what did not, and what we learned selling on x402 ourselves.
It adapts the seventh chapter of *The x402 Economy* and the measurement work behind How big is the x402 market? We run a seller, so we have a stake in the answer. We report our own results as they were.
What is for sale on x402, and at what price?
The best public view of x402 sellers is Coinbase's Bazaar catalog. It is incomplete, because it lists an endpoint only after its first payment settles through Coinbase's facilitator, but it shows each seller's price, description and recent usage. The Group read the whole catalog on 27 August 2026.
Prices are tiny. Of 14,781 active resources, the median price was one cent, the middle half of prices ran from $0.003 to $0.05, only two were free, and the most expensive was $1,000.
Revenue is concentrated. One seller of gift cards, which lets a buyer spend crypto at ordinary shops, earned about $21,000 of the catalog's $30,600 in thirty days. The ten highest earners took 78%. The remaining sellers averaged under fifty cents a month.
The most popular services are thin. The resources with the most distinct payers were simple lookups on the Ethereum blockchain: the balance of an address, the owner of a name, the current block number, the details of one transaction. Each drew about 1,000 payers and about 1,000 calls.
What is "calls per payer", and why is it the best early metric?
A useful signal for any pay-per-request service is the number of calls each paying customer makes. The Group calls it calls per payer.
- A ratio near one means each buyer tried the service once and left. That is the pattern of sampling: a developer testing an integration, an agent exploring what exists, a demonstration.
- A ratio of tens or hundreds means buyers came back. That is the pattern of a service built into someone's workflow, where the payments recur because the need recurs.
The popular blockchain lookups in the Bazaar had ratios near one. Many payers tried them; almost none returned. They were popular as samples, not as products. For a seller, calls per payer is a more honest early signal than revenue or payer count: a hundred payers who each called once is a demonstration; ten payers who each call daily is a business. Count your own tests separately, because early payers are often the seller, its friends and curious developers.
What happened when the Group sold an ephemeris on x402?
The Group sold two products over x402 from July 2026. Both computed the positions of the planets for a given moment. The cheaper tier, at $0.04, returned planetary positions, the moon's phase and the major angles between planets. The precision tier, at $0.32, used NASA's Jet Propulsion Laboratory ephemeris (DE440) and returned a fuller set of astronomical and astrological values with a cross-check between two calculation engines. (Both are still on the market today.)
The service went live on 22 July 2026. The Group paid it twice itself to test both tiers. Over the next four days two strangers paid it, one of them three times. Then payments stopped for more than a month. Our records disagree about whether one more payment arrived in late August, and about the exact total, but they agree that the total was about fifty cents.
The first explanation we reached was discovery: nobody could find the service. The Bazaar lists an endpoint only after a payment settles through Coinbase's facilitator, and this one never appeared in the catalog, even after it had been paid. That looked like a closed loop: no listing, so no buyers, so no listing.
The records showed that explanation was incomplete. Strangers had found the service and paid for it. The question was not how to get a first customer but why the customers did not return. For the cheaper tier the answer was plain:
- A free service covered much of the same ground. It computed similar positions with no payment at all, and it was reachable through the same channels an AI assistant would use to find tools.
- Conventional astrology services priced by subscription gave away a thousand or more calls a month before charging.
At four cents a call, the x402 tier was competing with free. There was a deeper problem: the product was built for an AI agent that pays mid-task, and very few agents have a task that needs a horoscope. The precision tier had a better case, because nothing free matched its accuracy, but it was labeled as astrology. The same calculations sold as positional astronomy might serve navigation, observation planning or solar-panel siting, buyers who value accuracy and do not look for horoscope tools. The market now lists the same calculations under astronomy routes (/v1/astronomy/positions/{date} and /v1/astronomy/precision-positions/{date}) as well as ephemeris routes.
What are the lessons?
Check the nearest free substitute, not the nearest paid competitor. A machine buyer compares a price with the cost of getting the same answer elsewhere, and elsewhere is often free. Before building distribution, know whether a free tool or a generous free tier already answers the same request.
Sell what an agent needs in the middle of a task. The services with recurring demand are those an automated process keeps needing: current data, computation it cannot do itself, verification, and access to something gated. A product people enjoy but software never needs does not find machine buyers, however easy it is to pay for.
Measure calls per payer, and remove your own tests.
Distribution is necessary, not sufficient. Listing in catalogs and directories helps buyers find a service. The Group's gateway was accepted by an independent directory, 402index, in September 2026, but a listing cannot create demand where none exists.
Expect concentration. In the Bazaar, a few sellers earned nearly everything. That is common in young markets and in platform markets generally. A new seller should plan for the median outcome, which is small.
What about publishers and the crawler market?
Not every x402 seller runs an API. Cloudflare's Monetization Gateway, announced on 1 July 2026, lets any site behind Cloudflare charge per request for pages, datasets, APIs or AI tool calls, with payment checked at Cloudflare's edge and settled in stablecoins through x402. Cloudflare's own examples of prices included one cent per request, a few cents per web search, and a tenth of a cent plus one cent per megabyte.
This changes who the sellers are. An API seller must build something an agent wants. A publisher already has pages that AI crawlers fetch, whether or not the publisher wants them to. For publishers, x402 offers a way to turn traffic that currently costs money into traffic that pays. Market analysts in August 2026 pointed to the gateway as the most likely source of new x402 demand in late 2026. Whether crawler operators pay rather than go elsewhere is not yet known.
What does the Group's market do differently now?
The Group's market at market.561.group has since grown to 130 offers. Three features of the market bear on these lessons:
- Many small, standards-exact operations at fractions of a cent, such as the 69 data conversions at $0.0012, are the kind of thing an agent needs mid-task and can't always run locally. We do not know yet whether they will find buyers.
- Refusing bad input before charging, so a buyer's first failed call costs nothing.
- Treating demand as unknown, not zero. The market's own README tells its operators that a broken demand sensor, missing counterparty receipts or unmeasured volume means demand is unknown, and says an internal self-purchase is not external demand. We do not report revenue here because we have no independently verified figure to report.
If you are deciding whether to sell on x402, the book's checklist is: find the nearest free substitute; choose a task where software needs the answer mid-flow; instrument calls per payer from day one; list in more than one directory; and budget for a small outcome. Then read x402 security for sellers, because a pricing mistake that lets a request skip its payment costs more than a slow month.
Where does this fit in the bigger picture?
x402 in late 2026 is a working rail with a small market. The book's final chapter lists seven conditions for growth, each with something observable to watch: services buyers return to (watch calls per payer), buyers trusted with budgets, honest measurement, fixed weak points, recourse, settled law and crawlers that pay. See also What is x402? and Who built x402 and who governs it.