The services reviewed: seven x402 services and their measured prices

x402.review reviews the x402 services our agents call, and the technical ground they stand on. Seven services are covered, four of them selling today. Every price on this site is measured from the live 402 handshake, never copied from a catalogue.

What this site is

Each review takes one service and answers four questions: what a call costs, what arrives when it is paid for, how an agent should use it, and where it stops being the right tool. The subject is narrow by design. x402 charges for a single HTTP request, so the useful questions about a service are concrete — does the endpoint answer, what does the challenge ask for, what does the response contain, and how does the amount move as the work changes.

Three longer pieces cover the ground beneath the reviews: what x402 is as a payment exchange, what an agent built to pay per call looks like from the inside, and what it takes to let that agent spend money without losing control of the bill. Read those first if the mechanism is new. The reviews assume it, and assume the reader is deciding whether to wire something up rather than browsing.

Promotional in intent, honest in content: the services reviewed here are sold through x402.press, so each piece is written to make that purchase easy to take with the facts in hand, including the facts that argue against it. A review that lists only strengths is an advertisement, and an advertisement is not worth the time it takes to read.

The seven services

Table 1 is the index. State, price and scheme for each service, with the review beside it.

Table 1. The seven services reviewed on this site, their state, their exact price on Base for a single call in USDC, and whether a metered scheme is offered. Source: live 402 handshakes decoded 30 October 2026.

Three of the four sellable services are chat completions, so the choice between them turns on the model you need, the amount you pay and where the machines sit rather than on what the endpoint does. HyperFusion quotes the same $0.00700 on Base as DeepSeek and serves a different set of models from GPUs it domiciles in the UAE. SpiralStake is the odd one out: no model, no completion, an unsigned transaction returned for a wallet to sign.

Three things need saying about how to read a row.

Sellable means the endpoint answers a paid call today. In build means it does not: there is no endpoint to call and no price to quote, which is why those three rows carry no amount.

The exact column is one chain out of seven — Base, in USDC, for a single call on a short prompt. Every sellable service also quotes Arbitrum, Polygon, Avalanche, Ethereum mainnet, Robinhood and Arc. On all four, Avalanche is the cheapest chain and Ethereum mainnet the dearest, and the gap between the two is measured in orders of magnitude rather than in margins. Base is the reference here because it is the default, not because it is the cheapest.

Metered records whether the challenge offers upto alongside exact. Where it does, the upto ceiling is a limit on what the call can cost at the max_tokens you authorise, not a second price. SpiralStake offers exact only, so the amount in its challenge is the whole price, paid before the work.

How a review is done

Every review starts with an unpaid request. Nothing is bought in order to discover a price.

An unpaid call to a paid endpoint returns HTTP 402 with a challenge: the scheme, the network, the asset, the amount, and a timeout. That response is decoded and recorded, and it is the only source this site treats as authoritative for what a call costs.

From there the price is measured rather than listed. A published price list, an /about page or a catalogue entry is used to check a figure, never to supply one. All three go stale in their own way, and none of them is the number the endpoint will demand at the top of the exchange.

Metered ceilings are reported against the max_tokens that produced them. An upto amount without its output allowance means nothing, so each ceiling is re-read at several allowances and published with the one that applies.

Claims about behaviour are tested on the endpoint where one exists — the free metadata routes, parameter pass-through, error paths — and reported as tested. Where no endpoint exists, the piece says so in its first line and describes a specification rather than a measurement.

Each review closes with a method note stating what was measured and when, and with what was not tested. Uptime, latency under load and behaviour under a failing facilitator are not measured here, because measuring them needs traffic this site does not generate. Where that gap matters, the piece says so rather than guessing at it.

What the prices say

Price on x402 is not a property of the work. It is quoted per chain, per scheme, and under upto per output allowance. Table 2 prices one call on three chains, and the spread is not subtle.

Table 2. One deepseek-v4-flash completion, short prompt, exact scheme, USDC, three chains. Source: measured 30 October 2026.

The same completion costs $0.002 on Avalanche and $3.00 on Ethereum mainnet: 1,500 times as much for an identical answer from an identical model. Against Base, the mainnet multiple is 428. Nothing in the request changed. The amount follows the chain the payment settles on, which makes the choice of chain a purchasing decision and not a technical footnote.

The metered ceiling is the second trap, because it looks like a price and is not one. deepseek-v4-flash on Base quotes an upto ceiling of $0.00955 at max_tokens: 1000, and $0.160787 with the field left out — nearly seventeen times the ceiling, from one omitted field, on a prompt that has not changed. Under upto the charge settles on tokens measured after the work, so neither figure is what the call costs. The ceiling is the most it can cost.

This is why the site exists. A price list cannot answer what a call costs, because the answer depends on four things at once: chain, scheme, prompt band, authorised output. Only the challenge carries all four, and reading it is free.

Live and in build

Four services here are sellable: DeepSeek, Qwen, HyperFusion and SpiralStake. Each answers a paid call now, each is priced on seven chains, and each has a review of a working endpoint.

Three are not. Wolfram Alpha, ElevenLabs and Mage are in build: no endpoint, no price, no date. Their pieces are previews — they describe what has been specified, from the request shape to the unit of billing, and they are marked as previews from the first paragraph. They are not reviews of a working service, and no amount on those pages is for sale.

The difference between the two kinds of piece is structural, not a matter of tone. A review of a sellable service carries price tables per chain, a measured ceiling against the max_tokens it belongs to, an example of an agent using the endpoint, and a section on where the service stops being the right tool. A preview carries the request shape, the billing unit and the design decisions behind them, each marked as specified rather than observed.

The catalogue lists 19 services. Four are sellable and reviewed here. The other fifteen are pending review: they are not covered on this site, and this page makes no claim about what they cost or how they perform.

Where to go next

• The mechanism. The technical pillar, What x402 is, walks the whole exchange — the challenge, the signature, the facilitator, the settlement — including the places where it fails.
• The spending system. How an agent pays covers the float, the three ceilings, the record, and what to do when a payment times out.
• The marketplace. https://x402.press/services/ — the index of services, sellable and pending.
• The catalogue. https://x402.press/catalog.json — the same list in machine-readable form, for an agent to read rather than a person.
• The studio. https://studio.x402.press/design — the design conversation behind the services, where the specifications described in the previews are worked out.

Read in this order if the mechanism is new: the pillar, then the review of the service you intend to call, then the challenge your own endpoint returns. The last of those outranks everything on this site, because it is the only one that states today’s price for your call.

Method note

Every price, ceiling and state on this page was captured on 30 October 2026.

Measured means read from a response: the amount in a 402 challenge, the token allowance an upto ceiling belongs to, the body of a paid call. It does not mean taken from documentation, a price list, a dashboard or a catalogue entry, and it does not mean inferred from any of them. Where a service does not yet answer a paid call there is nothing to measure, and the piece describes a design instead, labelled as such.

Amounts move. They move with the chain, with the scheme, with the prompt band a call falls into, with the max_tokens you authorise, and on some services with the hour — DeepSeek’s amounts rise inside its published peak windows, and the figures here were read outside them. The challenge in front of your agent is the price. These tables are what the price was when it was read.