What x402 endpoints actually charge vs what registries claim
A measured comparison of catalogued prices against the price the endpoint really quotes, across 4,251 routes where both figures exist.
DIRECT ANSWER
Across 4,251 routes where a catalogued price and a decoded live 402 quote both exist, 4,065 agreed and 186 did not, a drift rate of 4.4%. Of the disagreements, 100 were major and 47 extreme. The worst observed case listed a route at one dollar while the endpoint quoted one tenth of a cent, a factor of one thousand.
Key takeaways
- Roughly one in twenty-three listed prices disagrees with what the endpoint asks for.
- Disagreements are not evenly small; the tail includes hundred-fold and thousand-fold errors.
- Only the live 402 requirement is authoritative at purchase time, whatever a catalogue says.
Two prices exist for the same route
Every listed x402 resource carries a catalogued price, which is what a registry entry says the route costs. Separately, when a strict unpaid check reaches the exact advertised route and receives a decodable HTTP 402, that response states what the endpoint is actually asking for right now. These are different facts recorded at different times by different parties.
Most of the time they agree, which is why the disagreement is easy to miss. The comparison only exists where both are present: a catalogued figure to compare against, and a confirmed quote that decoded cleanly to a dollar amount. In the current corpus that intersection is 4,251 routes.
What the comparison found
Of 19,259 confirmed quotes, 19,257 decoded to a usable dollar amount and 2 did not, an undecodable share of about one in ten thousand. Of the decoded set, 4,251 had a catalogued price to compare against and 15,006 were live-quote-only, meaning the endpoint told us a price no registry had recorded.
Within the comparable set the agreement bands are heavily weighted toward match, with a long and consequential tail. A minor difference is the kind of rounding or currency-precision gap a buyer would not notice. Major and extreme differences change the economics of a call.
| Band | Routes | Reading |
|---|---|---|
| Match | 4,065 | Catalogue and live quote agree |
| Minor | 39 | Small difference, unlikely to change a decision |
| Major | 100 | Difference large enough to change unit economics |
| Extreme | 47 | Order-of-magnitude or worse disagreement |
The worst cases run in both directions
An Alchemy token-price route was catalogued at one dollar per call while its live quote asked for one tenth of a cent, a thousand-fold overstatement in the catalogue. A buyer trusting the listing would have priced a workflow at a thousand times its real cost and probably rejected it.
The reverse also occurs. A route catalogued as free that quotes a small non-zero amount will break an agent that budgeted nothing for it. Neither direction is a scandal on its own, but both mean the same thing operationally: a catalogued figure is a hint, and the requirement returned at request time is the number that governs.
- Catalogue high, quote low: workflows priced out that were actually affordable
- Catalogue low, quote high: budgets exceeded and calls failing mid-run
- Catalogue free, quote non-zero: agents with no payment path stall
- Catalogue present, route no longer 402: the price is moot
Why catalogues drift
Registry entries are usually written once, at registration, and rarely revisited. A provider who repriced a route six weeks later has no obligation to update every catalogue carrying it, and in most cases no mechanism to do so. Drift is the expected steady state rather than a sign of neglect.
Templated route families make it worse. One registered entry can stand for hundreds of concrete routes that were never individually priced, so a single catalogued figure gets attributed to routes that may have diverged. This is also why any pricing analysis has to collapse route families before taking an average.
How the comparison is guarded against false alarms
A drift measurement is only useful if it fails loudly when the measurement itself breaks. The check refuses to report healthy if more than 15% of confirmed quotes stop decoding to a dollar amount, or if the extreme-drift share crosses 5%, or if the underlying checks go stale past fourteen days.
Those thresholds are deliberately loose. They are set to catch a decoder regression or a crawl that quietly stopped refreshing, not to police normal repricing. The current run sits far inside all three, with an undecodable share of 0.01% and no stale checks.
What a buyer should do with this
Treat a listed price as a filter, not a commitment. It is good enough to shortlist candidates in a capability and bad enough that no budget should be built on it without a check. The unpaid 402 request that returns a requirement is cheap, needs no payment, and settles the question directly.
For a workflow that will run at volume, check the quote for the exact method and URL the agent will call, not a sibling route on the same host. Price is attached to a route, and a family that shares a title can carry different amounts.
- Shortlist on catalogued price and capability
- Confirm with an unpaid request to the exact route the agent will call
- Re-check before any change in call volume
- Budget from the quoted requirement, never the listing
What a provider should do with this
If your listed price and your quoted price disagree, buyers comparing you against a competitor are comparing the wrong number. Correcting the registry entry is usually a small change with a direct effect on how your endpoint is evaluated.
The more damaging case is an endpoint whose requirement will not decode at all. Two routes in the entire corpus fell into that category this run, and each one is invisible to any client that reads prices programmatically. A requirement that a strict parser rejects is a defect worth fixing even when a tolerant client happens to cope.
Limits of this measurement
Every figure here describes published prices and observed requirements. None of it is evidence of demand, and a confirmed quote is not a settled payment. A route can quote a price accurately and still return nothing useful once paid.
The comparison also cannot speak for the 15,006 routes that quoted a price no catalogue recorded, or for the far larger set that never presented a decodable 402 at all. Drift is measured where both numbers exist, and that intersection is a minority of the corpus. Figures are as of the 5 August 2026 refresh and are appended to a dated history so the trend becomes measurable rather than anecdotal.
Related directory entries
Sources and methodology
TOLL·402 distinguishes public claims, registry discovery, unpaid quote checks and settled paid-call verification. Sources below support the visible claims; presence in a registry is not treated as verification.
- x402 rate card — Provider-weighted price benchmarks built from the same price evidence.
- TOLL·402 resource explorer — Per-route catalogued price and quote outcome for every canonical resource.
- x402 protocol documentation — Asset and decimal handling used to decode a requirement into a dollar amount.
- TOLL·402 discovery corpus — Downloadable records behind the comparison.