2026-10-14 · by minia2a
On 2026-10-14, a Chainalysis report — covered the same day by Fortune and PaymentsJournal — put a number on something the x402 ecosystem had only been able to argue about: more than 138 million discrete payments settled through x402 on Base in Q3 2026, up from roughly 1,500 in mid-2025.
Over the sixteen months leading up to September, more than 87% of these payments fell between one-hundredth of a cent and $0.10.
But the count is not the most interesting line in the report. This is:
That is the whole story compressed. In late 2025, an agent wallet sat funded for two and a half months before it paid anything. By Q3 2026 the same median is under twenty minutes, and roughly two-thirds of new wallets spend within an hour of receiving funds.
A payment rail can carry a large transaction count and still be a science project. What separates a rail from a demo is whether the funded wallet comes back. On that question the report is blunt: around 90% of qualifying active wallets received another deposit within 30 days, up from roughly 71% a year earlier.
Read those two numbers together and the shape of the thing changes:
For anyone publishing a per-call endpoint, the implication is specific. A client that pays within an hour of funding cannot be asked to complete a multi-step onboarding flow first. The first 402 has to be the onboarding — the challenge itself has to teach the client how to pay it. If your first response is a landing page and a sign-up form, you are optimising for a user who no longer exists.
The Chainalysis categories are worth listing plainly, because they are more mundane than the "agents buying things" framing suggests — and mundane is what makes them real:
| Category | What it looks like on a 402 |
|---|---|
| Model inference | Per-call completions, described as one of the largest categories |
| Code review / analysis | Static and security checks billed per invocation |
| Financial intelligence | Prices, on-chain reads, market data |
| API services | Lookups and transforms wrapped as billable calls |
| Gaming & gambling | Per-action endpoints |
| Token issuance | Deploy-and-return flows |
None of these require a human to be in the loop, and all of them are priced below the point where a card rail makes sense. That is the mechanical reason the sub-ten-cent share is 87% and not 8%: for a two-cent API call, the fixed cost of the traditional payment path is the product.
One day earlier, on 2026-10-13, Ripple and Privy — a Stripe company — announced programmable wallet infrastructure embedded in the XRP Ledger, letting developers create and drive XRPL accounts programmatically so an agent can hold and move XRP, RLUSD and other native assets. Each new account needs a 1 XRP minimum reserve. Spend-limit policy controls and gas sponsorship are described as in development and not yet generally available; the current release covers core transaction signing.
The useful part is not the announcement — it is the list of what is still missing. An agent that can sign a transaction is not yet an agent that can be trusted with a budget. Spend limits are the difference between "the agent has a wallet" and "the agent has an allowance." Until that ships generally, per-call services on a new chain inherit an unbounded-spend problem they did not create.
Two caveats belong next to every number above, and the coverage states them:
This is the same discipline a per-call market has to apply to itself. A cumulative counter can only go up; it wakes up the numbers of every probe, every test script, and every trial that never repeated. The honest question is not "how many calls did the rail carry" but "how many distinct payers came back."
The report's most transferable finding is the repetition rate: ~90% of active wallets deposited again within 30 days. Repeat funding is the one metric that cannot be inflated by a scanner, because a scanner does not care whether you come back.
Any operator of a per-call market should be able to print the same number for their own ledger. If they cannot, the cumulative counter they are printing is measuring traffic, not customers.
Applied to us, honestly: minia2a publishes 1,696 catalog endpoints that settle in USDC on Base, with five free trial calls per signed wallet and no registration. The cumulative x402 figures in this post are the ecosystem's; our own settled-external-payment count is four figures' worth of orders of magnitude smaller, and we quote it as a live read rather than bury it, because a number that only moves one way is the number to distrust.
The direction of travel is nonetheless clear and it is the useful part. A rail where the median wallet waits 73 days is a rail you build demos on. A rail where the median wallet waits 18 minutes, and 90% of wallets fund a second time, is a rail you build products on. The second one arrived sometime this quarter. Everything below ten cents is now a market, not a proof of concept.
minia2a is an x402 pay-per-call API market for agents — 1,600+ endpoints, USDC on Base, five free trial calls per signed wallet.
Sources: the Chainalysis x402/Base figures as reported 2026-10-14 by Fortune (via Yahoo Finance) and PaymentsJournal ("Agentic Commerce Transactions on x402 Protocol Are Picking Up", Wesley Grant); the Ripple/Privy XRPL wallet announcement of 2026-10-13 as reported by Forkast, including the cited XRPL (~13–14M agent payments) and cross-chain (191.75M) totals and the 25–30% test/leaderboard-noise estimate attributed to BigGo Finance. First-party figures (1,696 endpoints, USDC on Base, trial policy) are read from the live service, not estimated.