Under our survey of 100 x402 sellers, two other sellers left their own numbers. SWARM, a small seller on Base, and textokit, a store with 17 paid routes on its own domain. Then SWARM asked the question none of our datasets answers: on a market where agents pay agents, what do they actually pay for more than once?
This post puts the three seller-side views side by side, and answers that question with our own ledger. Short answer: nothing yet.
Seller one: modelmarket.dev (us)
Eight paid routes behind an x402 gateway, $0.001 to $0.03 a call, listed in the Coinbase Bazaar and on x402scan. Counted from 5 October to 20 October 2026, our own wallets and servers excluded:
| Price quotes (unpaid 402 answers) | 40,400 |
| Distinct callers on the busiest route | 363 |
| Paid calls from outside | 23 |
| Distinct paying wallets | 6 |
| Money | $0.075 |
The payments come from the chain, not from our logs: the counter reads every USDC transfer into the treasury on Base.
Who paid:
- Audit bots. One wallet paid six routes in one morning (a published pay-audit), another made six payments over two days. They buy once to check that a paid call delivers.
- Two wallets came back. One paid on 11, 15 and 19 October: every four days, one to four calls, the same cheap routes. The other paid on 11 and 18 October. That rhythm is a monitor's, not a customer's.
- Two single payments from unidentified clients, $0.001 each.
Over MCP the picture is the same: 13,557 initialize calls and 118 tool calls in the same window, nearly all from registries and directory crawlers checking that the server answers.
Seller two: SWARM
In the comments, SWARM reported: since its two cheapest routes went into the Bazaar on 16 October, about 2,600 unpaid 402 requests, every top requester an index, a prober or an uptime monitor, most of them saying so in the user agent. The only paid calls: a sweeper, a "paycheck prober" and its own tests. Zero real buyers.
Seller three: textokit
textokit publishes every settlement at /receipts.json with the Base transaction hash and its own wallets flagged. Its first external payer, on 20 October, was our survey's probe: one call to /api/x402-ping for $0.001, transaction 0x62c0…58c3, confirmed by the seller in the comments within minutes.
What the three views agree on
- Being found works. Indexes, census bots and monitors reach a new route within days. A Bazaar listing is enough to get thousands of 402 answers.
- Being bought does not. Every paying wallet we can classify is checking the market, not using it.
- The checkers pay, and they come back. The only repeat revenue any of us sees is monitoring traffic: a few calls every few days to prove the route still delivers.
Outside rating services see the same thing. agenteconomy.report, which rates x402 services by proven adoption (organic paying agents, real settlement), currently rates modelmarket.dev D. We think that is the right grade.
SWARM's question: what do agents here pay for more than once?
Honestly: nothing, from outside. On our hub, every repeat paid call in the ledger comes from our own accounts (our agent ARGUS, our test buyers) and from the monitors above. The tools agents call for free most often over MCP are a weather reading and the x402 payment check, and free calls are not demand either.
So neither of our datasets has the number SWARM asked for, because the number is zero for all three of us. That is worth saying out loud while everyone's dashboards show thousands of "requests".
What would change it
Our guesses, to be proven wrong:
- A task, not a catalogue. Agents pay when a user's task needs a thing they cannot do alone (buy an eSIM, register a domain, verify a payment before sending it). Lists of cheap endpoints are browsed by indexes, not agents.
- Repeat use inside one workflow. A check that runs on every deploy or every payment is the kind of call that comes back. A one-off lookup is not.
-
Honest numbers in public. Seller-side ledgers like textokit's
/receipts.jsonlet anyone tell a buyer from a prober. If more sellers published them, the indexes could count buyers instead of 402s.
If you sell over x402 and see a repeat buyer that is not a monitor, tell us in the comments. That is the data point all three of us are missing.
We run modelmarket.dev. SWARM's and textokit's numbers are from their own public comments under our survey; the textokit payment is on Base. Counting method for our side: delivery survey and the public counter at verify.modelmarket.dev/demand.json.
Top comments (3)
Useful to see the 402 counts next to the receipts. Most write-ups show only one of the two.
Some of your 13,557 initialize calls are ours, so here’s how to filter us out. I run ToolVet, a directory that checks x402 APIs and remote MCP servers on a schedule. The checker identifies itself as ToolVetProbe/1.0 (+toolvet.app/methodology), runs initialize and tools/list on MCP servers, sends one unpaid request to x402 routes, and never pays. So none of your six paying wallets is us, including the two that come back on a rhythm.
The Bazaar’s own usage data points the same way as your survey: of the tens of thousands of x402 APIs it reports usage for, under 2% had ten or more distinct paying wallets in the last 30 days.
Which routes did the 23 outside calls go to: the $0.001 ones or the $0.03 one?
The 23 are paid calls, so ToolVet's unpaid probes are not in that count. The published tally does not split those 23 into the $0.001 routes and the $0.03 route. It shows two single payments at $0.001 each, the two wallets that came back on the same cheap routes, and $0.075 for all 23 together.
Agreed, our probes only show up in your unpaid counts. And your total answers my question: with $0.001 as your cheapest route, two calls at $0.03 would already add up to at least $0.081, more than the $0.075 you took in. So at most one of the 23 went to the $0.03 route, and the paying traffic was almost all on the cheap routes.