The State of x402

How the index breaks down

Grade distribution across all 3,503 active services.

Most of the catalog is usable. Of the services we can currently probe, 79% grade A or B — an agent can discover them, pay them, and parse what comes back. But nearly one in five (18%, or 620 services) grades D or F: broken enough that an autonomous agent will struggle or fail. Another 972 listings sit inactive — advertised in a directory but not returning a valid probe when we reach them.

Of the graded catalog, 2,794 services (80%) are callable through the gateway — graded D or better and priced at or under the $0.25-per-call ceiling. The rest stay listed and graded, just not payable through us.

Where a service comes from predicts whether it works

Share grading D or F, broken out by the directory each service was discovered in.

This is the clearest signal in the whole index. Coinbase's curated x402 Bazaar supplies the bulk of the catalog and is remarkably healthy — only 8% of Bazaar services grade D or F. Pull from an open, self-submission directory instead and the picture inverts: 64% of the services listed on 402index.io are broken for autonomous use. A service discovered through the open directory is roughly eight times likelier to be unusable than one from the curated source.

The takeaway for anyone building an agent: the directory you discover a service in is a quality filter in itself. An open, unmoderated list is mostly noise; a curated one is mostly signal. Grading exists to catch the exceptions in both directions — the broken Bazaar service and the working one nobody vetted.

Thousands of vending machines, a handful take the coins

Distribution of 292,967 paid calls over the last 30 days.

The catalog is broad, but real usage is not. Across the 30-day window we recorded 292,967 paid calls spread over 2,784 services that saw any demand at all. Yet the top 10 services absorbed nearly 80% of every call (233,077 of them), and the single busiest domain — x402.twit.sh — took 41% of all traffic by itself. At the other end, 81% of the services with any demand saw fewer than 10 calls in the whole month, and 761 saw exactly one.

It is an economy of a few busy machines and a very long tail of ones that have been paid once, to prove they can be.

Calls and payer counts aggregated by domain over the trailing 30 days. "Distinct payers" is the number of separate paying wallets we observed.

Demand doesn't track quality

Two rows in that table tell the more interesting story. The third-busiest service, agents.chain.link, grades D — nearly 29,000 calls flowing to something our probe rates poorly for autonomous use, from just two payers. When a caller hardcodes around a service's gaps, it can drive enormous volume to an endpoint a fresh agent could never call unattended. Volume is a measure of who already integrated, not of whether the next agent can.

The inverse shows up too. Further down the list, api.onesource.io logged about 1,900 calls from 1,452 distinct payers — the broadest paying audience in the index, roughly one call each. That is the opposite shape from x402.twit.sh's 120,000 calls across 103 payers: one is a service many agents try once, the other a service a few agents lean on hard. Grades describe the payment surface; demand describes the relationships built on top of it. This report is the first time we can watch both move.

What agents actually pay

Per-call price across the 3,503 active services.

x402 is a micropayments protocol and the catalog prices like one. The median call costs a single cent. About three-quarters of priced services charge a nickel or less, and 93% sit at or under the gateway's $0.25 ceiling. Only 210 services — 6% of the index — price above $0.25 per call, and those are the ones our gateway won't front payment for.

Bar length is share of the 3,503 active services in each price band.