The state of x402 in mid-2026

On 2026-07-14 the Linux Foundation announced the operational launch of the x402 Foundation, with Coinbase contributing the protocol and roughly 40 members, 17 of them premier. An executive-director search and a technical steering committee are underway.

Over the 30 days to mid-July 2026 the protocol carried on the order of 24 million dollars across about 75 million transactions, roughly 94,000 distinct buyers and 22,000 sellers, at an average payment near 32 cents. That is the thesis working as designed: charges too small for a card network to process, clearing in stablecoins between software agents. It is also a rounding error next to what any premier member settles in a day, so the honest read is early traction, not scale.

Sellers list services in several registries: the permissionless 402 Index, x402scan (Base and Solana, with an embedded-wallet test flow), x402-list, and the Coinbase CDP Bazaar, which agentic buyers query directly. The catch is liveness. A July 2026 verification by 24K Labs probed all 22,545 services in the CDP Bazaar and found only 5,792 returned a valid 402 response, a 74 percent dead-or-misconfigured rate. That gap is why verified-working directories and on-chain payer counts, the one signal a listing cannot fake, are becoming the trusted layer.

Coinbase documents two rules that explain most of that gap, and they also define what our own counts are counts of. A resource is added to the Bazaar only when it first settles a payment, so verification alone never lists you, and nothing that has failed to earn once is in there at all. A resource is then dropped once it has gone 30 days without activity. So a listing means the service took money at least once and has done something in the last month. A service can be dead for four weeks and still listed, which is room enough for most of a 74 percent miss rate. It also means every count on this site is a count of survivors, and the true population of x402 services is larger than the directory and earns less than it, not more.

The volume figure above is the whole rail. What reaches a seller is a different and much smaller number, and it is the one that decides whether to build here. Across 14,064 priced services we counted 339,178 paid calls in 30 days, worth roughly $9,747. The median listing earned three cents for the month, and 12 listings in the entire economy cleared $100.

Each stage is a subset of the one above it, and the percentage is a share of all 14,064 counted listings. Revenue is estimated as price times Coinbase's own 30-day call counts, and Coinbase does not document how it counts, so these are its numbers reported faithfully rather than ours. The population is the survivor set described above.

Facilitators such as the Coinbase CDP facilitator and PayAI settle both flows. Client stacks are split, so a service that speaks only v1 is invisible to the newer agent wallets, and one that speaks only v2 is unreadable to legacy index probes.

Supply is not the constraint and neither is price; the average transaction is a third of a cent to a few cents, and the shelves are full of extraction, metadata, and audit endpoints. The scarce resources are discovery and trust. The services that break through tend to publish a verifiable track record, answer both protocol versions, and carry a story a human will actually share. For a brand-new endpoint, a single credible human referral is worth more than another registry row.

Method note: figures are reported as approximate because they come from third-party trackers and a point-in-time probe; a paid brief states the same caveats. This sample is AI-produced and labeled as such, like every Circadian deliverable.