Paid Calls Nearly Tripled. Two Sellers Carried Half of Them.
October 2026 edition · every figure comes from one read of the catalog and one check of the fleet, both dated October 2, plus an on-chain sample and a Tempo read the same day · published monthly · data free to download
This is the first edition of the Orcpin Agent Payments Index: the same measurements, taken the same way, every month, so that a number here can be compared with the number a month ago. Every figure on the page comes from files anyone can download, and every figure is reported twice where we can: as the catalog lists it, and as it looks once you check it.
This month the two versions disagree more than ever. Reported paid calls across the x402 catalog rose from 281,361 to 789,704. Half of them went to two sellers.
The month in four numbers
789,704
paid calls reported across the catalog, 30 days
September: 281,361
50%
of those calls went to the two busiest sellers
the busiest alone: 38%
732
services paid by six or more different wallets
3% of 22,243 listed
$1,388
paid to Tempo services in 30 days, verified on chain
from 3,061 paying wallets
The Index, October 2
Measure
September 1
October 2
Services listed (x402 catalog)
14,913
22,243
Distinct sellers
1,127
1,495
Services listed since September 18 · gone since
not measured
10,389 · 3,771
Paid calls reported, 30 days
281,361
789,704
Paid calls with the busiest seller set aside
not measured
490,457
Share of calls to the busiest seller · two busiest · ten busiest
not measured
38% · 50% · 80%
Spend claimed at listed prices, 30 days
$32,208
$50,894
Services paid by six or more wallets
see the wallet chart
732
Services paid by one wallet or none
74%
64%
Typical price per call
$0.01
$0.01
Prices raised · cut, since September 18
not measured
135 · 186 of 11,704
Sampled payments from established wallets
not measured
53%
Services in the daily check: healthy · failing repeatedly
The Short Version: More of Everything, Mostly in Two Places
The catalog grew again, and fast. It now lists 22,243 services from 1,495 sellers, up from 14,913 and 1,127 a month ago. Between September 18 and October 2 alone, 10,389 services were listed and 3,771 disappeared. The growth is concentrated: five sellers listed 50% of the new services, and all the new services together carry 5% of the month's reported calls. Most new listings carry almost no calls.
Reported paid calls rose from 281,361 to 789,704. That is the number dashboards will quote, and it needs its context: the busiest seller alone accounts for 38% of it, and the two busiest for 50%. Set the busiest seller aside and 490,457 calls remain.
Breadth barely moved. Only 732 services, 3% of the catalog, were paid by six or more different wallets in 30 days. Prices held still: the typical call costs $0.01, and of 11,704 services priced in both September 18 and October 2, 135 raised their price and 186 cut it.
Where the Calls Went
Paid calls in 30 days by category: September vs. October
Each listing is placed in a category by keywords in its name, description, and tags; the
rules are in the data file's method note. Prices and trading rose from
37,722 to 387,499
calls, and AI model calls from 17,532 to
113,422.
The two categories that grew the most are each mostly one seller. In prices and trading, the busiest seller carries 77% of the category's calls; in AI model calls, 86%. A category chart can't tell a market from a single busy seller, which is why the Index reports concentration beside every total.
Elsewhere growth was ordinary. Web search and scraped data went from 80,700 to 94,412 calls, with its busiest seller at 37%; blockchain data from 60,024 to 75,363, its busiest seller at 22%.
Who Is Paying: The On-Chain Check
The catalog counts calls and paying wallets, but it can't see who those wallets are. So for the 12 busiest sellers we could read on the Base chain, which together carry 77% of the catalog's reported calls, we read recent payments and sorted the most frequent paying wallets by how they behave. The rules are fixed and published with the data:
Broad testers paid 20 or more different sellers within a single hour: the pattern of directories and scanners that test-buy many services once.
New single-seller wallets received their first dollars within 30 days and send almost all their payments to one seller. That can be a loyal new customer, or one operator's own automated wallets; the chain can't tell the two apart, and we don't guess.
Other paying wallets are everything else: wallets more than a month old, or that pay more than one seller. It's the closest the chain gets to the pattern customers leave.
We read the latest 1,200 payments to those sellers and classified the 835 made by their most frequent payers, 76 wallets in all. 53% of those payments came from established wallets. 46% came from new single-seller wallets, and 0.6% from broad testers. Three of the fifteen busiest sellers settle on other networks and were not read.
Read that the narrow way. The busiest sellers' heaviest payers split close to evenly between established wallets and wallets that appeared in the last month to pay one seller. The second group is not proof of anything, and some of it may be new customers. But when a call count nearly triples in a month, how much of it arrives from wallets that are weeks old and pay only that one seller is the first thing worth knowing, and no catalog shows it.
We publish shares, never a verdict on a named seller. The point is narrower: a reported call count is a claim, and this is how much of the claim looks like customers when you check.
Every Listing in the Catalog
How many different wallets paid each listing in 30 days: September vs. October
64% of listings were paid by one wallet or none over 30 days, down
from 74% in September. Source: Coinbase's own per-listing count
of distinct paying wallets over the prior 30 days, read on October 2 and September 1.
Paid calls per paying wallet, for listings with at least two paying wallets: September vs. October
Dividing a listing's paid calls by its paying wallets separates repeat use,
where the same wallet keeps coming back, from one-time use, where many
wallets each pay about once. Whenever you see a count of paying wallets, the calls-per-wallet
figure belongs beside it.
The Services That Still Answer
What the daily check found: September 1 (a 494-service sample) vs. October 2 (the whole fleet)
Healthy means the service answered correctly at least five times with no
failures. Too few samples means fewer than five checks so far.
Flagged means it failed, kept failing, or stopped answering. From this edition
the check covers every service we track, 4,229 of them, four times a day; in
September it read a sample of 494. Definitions are at
orcpin.dev/methodology.
On October 2, 2,411 services were healthy. Among the flagged ones, 203 had failed several checks in a row and 26 did not answer at all; a service can carry both flags, so those two counts overlap and don't add up to the chart's flagged bar. These are recorded states, never a score.
Beyond x402: Tempo
Stripe and Tempo's service directory lists 137 services that settle on the Tempo network. Unlike the x402 catalog, it publishes no usage figures, so we read the payments directly. Behind the listed services we found 40 receiving wallets; 29 of them received at least one payment in the last 30 days, and the table counts the transfers into those.
Tempo, verified on chain
7 days
30 days
Paid to listed Tempo services
$241
$1,388
Payments
14,286
129,359
Distinct paying wallets
1,120
3,061
Receiving wallets that were paid (of 40 found)
25
29
The two numbers are not the same kind and shouldn't be added together. The x402 figure ($50,894) is what the catalog's reported calls would be worth at listed prices. The Tempo figure ($1,388) is money that actually moved, read from the chain. The gap between a claim and a payment is the reason this Index exists.
Orcpin's Own Visitors
Who visits Orcpin's paid services, as a share of daily requests: September vs. October
Watchers are crawlers, indexers, scanners and automated catalog reads.
Evaluators asked a paid service for its price and did not buy.
Purchasers paid; they remain under one in a thousand requests. Shares, never
counts: one seller's sales volume is private. October shares cover the 13 days ending
October 1; September's, the 13 days ending August 31.
How We Measured
Catalog: every page of api.cdp.coinbase.com/platform/v2/x402/discovery/resources on October 2, 2026, duplicate URLs removed (23,041 reported, 22,243 unique). Per-listing 30-day figures are Coinbase's own quality fields, copied rather than audited.
Change since September 18: the per-listing rows of the September 18 and October 2 reads compared by URL. September 18 is the first read whose rows were kept; the September 1 edition kept only totals.
Spend claimed: reported calls times listed price, leaving out prices above $1,000 a call. It overstates real revenue wherever a call was counted but never paid.
On-chain check: Base USDC only, the busiest sellers by reported calls, their most recent payments and most frequent payers, each payer classified by its own recent history. Sample sizes and rules are in the data file.
Tempo: the MPP directory's catalog feed, one unpaid request per listed endpoint to find each service's receiving wallet, then the transfers into those wallets from Tempo's public RPC.
Daily check: free test requests to every service we track, four times a day; the state of each on October 2 is pinned for this edition.
Orcpin's visitors: coarse classes only, kept as daily counts; no addresses, request details or identities are stored (/privacy).
One edition, one dataset: every figure on this page comes from the October 2 files, and nothing changes after publication. GET https://orcpin.dev/research.json shows the live state of the daily check, which moves every day.
Disclosure: Orcpin sells x402 services, including the reliability checks described here; this is measurement of a public system by a participant in it.
Orcpin reports counts, dates and dollars, with the rules stated. It never gives advice and never assigns a score. This report is measurement of a public system, not financial, legal or investment guidance. See /terms.
The Data
Everything behind this page is free to download, so any number here can be checked or recomputed: https://orcpin.com/data/latest.json lists the files for the current edition. They include the catalog summary, one row per listed service, the on-chain sample, the Tempo read, the daily-check pin and the visitor shares, each with its definitions inside.
Past Editions
Every edition keeps its own permanent page, exactly as published, and its data stays downloadable, so a figure cited from any month can always be checked.
CDP Bazaar discovery catalog, Coinbase Developer Platform: api.cdp.coinbase.com/platform/v2/x402/discovery/resources, read October 2, 2026 (and September 18 and September 1).
Base chain payments: Blockscout's public API at base.blockscout.com, read October 2, 2026.
Tempo and Stripe: the MPP services directory and its catalog feed, and Tempo's public RPC at rpc.tempo.xyz, read October 2, 2026.