We probe 517 paid x402 APIs every night. Here is what the first numbers say

AI agents that pay for APIs with x402 usually pick a seller from a directory listing. A listing tells you a service exists and what it costs. It does not tell you whether it answers, how fast, or whether anyone else pays it.

So I started probing them. Every night a small script on a Raspberry Pi takes the x402 origins in the Coinbase Bazaar with at least two paying wallets in the last 30 days (517 so far, about 500 per night), sends one unpaid request to a real paid endpoint, and records whether a valid 402 Payment Required challenge comes back and how long it took. It never pays. The results are a free table and JSON feed: factline.dev/x402/health and GET https://factline.dev/v1/x402/sellers.

This is after two nights, so read it as a first snapshot, not a trend.

What is in the sample

• 517 origins (hosts) with 2,555 paid resources between them, 4.9 per origin on average, 72 at most.
• Selection: origins with 2 or more distinct paying wallets in the Bazaar over 30 days. Payer counts include sellers paying themselves, so treat them as an upper bound.

1. Almost everything answers

	Night 1	Night 2
Probes	500	501
Valid 402 challenge	496 (99.2%)	500 (99.8%)

Over both nights, 513 of the 517 origins answered every time. 3 never answered, 1 answered once out of two.

That is better than I expected. The catch is the next section: answering is not the same as answering fast.

2. The sellers people pay are three times faster

Origin group	Origins	Average uptime	Median response
10 or more paying wallets	106	100%	501 ms
2 to 9 paying wallets	411	99.1%	1,609 ms

The slowest valid 402 took 26 seconds. The fastest took 139 ms. An agent that waits 26 seconds for a price quote before it has even paid is going to time out or give up, and the payer counts suggest buyers notice.

3. Calls are concentrated in very few hosts

Paying wallets (30 days)	Origins	Calls (30 days)	Share of calls
100 or more	5	466,991	70%
10 to 99	101	161,472	24%
2 to 9	410	40,817	6%

Five hosts carry 70% of the paid calls in this sample. Four out of five sellers share 6%. If you are building on x402 as a seller, the honest takeaway is that a listing alone does very little.

4. Base is the default, Solana is the common second

Networks offered	Origins
Base only	309 (60%)
Base and Solana	106 (21%)
Base, Solana, Polygon and Arbitrum	38 (7%)
Solana only	3
Base Sepolia (testnet) only	5

508 of 517 origins (98%) accept USDC on Base mainnet, and 181 (35%) also accept Solana. Five are listed with paying wallets but advertise only Base Sepolia, the testnet.

5. Most prices are under a cent

Lowest price per call	Origins
under USD 0.001	1
USD 0.001 to 0.01	294 (57%)
USD 0.01 to 0.10	206 (40%)
USD 0.10 to 1	10
USD 1 or more	5

What I would do with this as an agent builder

• Check that a seller answers its 402 quickly before you route traffic to it. A slow challenge is a cheap early signal.
• Prefer sellers that offer more than one network, so a wallet on either Base or Solana can pay.
• Do not read a high call count as trust. Look at distinct payers and response times together.

The data

• Table, filterable by uptime and payers: https://factline.dev/x402/health
• JSON for your agent, free: GET https://factline.dev/v1/x402/sellers
• Method: one unpaid request per origin per night against a listed paid resource; "ok" means a parseable x402 challenge (HTTP 402 with payment requirements). Nothing is paid, nothing is stored beyond status, time and date.

I run this as part of Factline, which sells official-records checks to agents over x402 (company checks, sanctions, supplier invoices). The health data is also behind a payee check that tells an agent whether to pay an x402 seller it has never used. But the table above is free, and I will post an update once there are 30 nights of data.

If you run an x402 service and want your origin added or removed, email [e-mail].