Pricing Intelligence · Collected

The Agent Pricing Index

products · categories · every figure read off the vendor's own page and dated OBSERVED
of these products route you to a sales call for at least one tier, and of those publish no price anywhere at all. Only of let you read a full price and act on it without talking to a human — and they are almost all model APIs and hyperscaler seats, not agents. The ones that still publish are quietly switching the thing they charge for — from a seat, to a unit nobody can count, to an outcome. This index records what each one actually charges, on the day it was read, with the quote that proves it.
Products
Sales call required
No price at all
Buy without a call

The Unit Problem — the 27 published unit prices on one axis, and what one unit of work costs once the vendor's own page does the arithmetic · Cite this index — license, fields, corrections, API.

Counts on this page are computed from the dataset every time it loads. None of them are typed by hand. The sales-gated figure includes every product that publishes no price at all — the two nest, they do not add.

Three things the table says

FINDING 01 OBSERVED

Outcome pricing stopped being a slogan and got a number

vendors now publish a per-outcome price you can put in a spreadsheet, and they disagree with each other by more than 2x on what an outcome is worth. Intercom's Fin is the only one that defines the unit tightly enough to forecast: one billable outcome per conversation regardless of how many actions the agent takes, and nothing charged when the conversation is handed to a human.

$0.90Gorgias / resolution
$0.99Intercom Fin / outcome
$2.00Agentforce / conversation
$1.8811x / lead contacted

The tell is what the loudest outcome-pricing marketer does. Sierra sells outcome-based pricing as its headline differentiator and publishes no number anywhere; its pricing page returns a 404. And Decagon, which offers buyers a choice between a per-conversation meter and a per-resolution meter, reports that most customers pick per-conversation. ANALYSIS Given the choice, buyers take the predictable meter over the aligned one.

FINDING 02 OBSERVED

The credit fog is the dominant model, and the meters sit apart

of the products in this index meter in a private currency — credits, actions, activities, messages — rather than in dollars or in anything the buyer can count in advance. Where the plan states both a price and an included unit count, the effective unit price can be derived. The spread across those derived rates is the finding.

$0.0012Make · per credit
$0.25Lovable · per credit
1,250xLindy · cost range of one task
6.9xRelevance · top-up vs in-plan

Two exhibits. Gumloop indexes a credit at $0.005 and then sells 20,000 of them for $37, because 12,600 are "bonus" — so the price it publishes and the price you actually pay differ by 2.7x, before an 8% orchestration fee that rises to 16% if you bring your own API keys. Lindy states that a single task costs between 2 and 2,500 credits, which means the buyer cannot forecast the bill by a factor of 1,250 even after reading everything published. ANALYSIS GitHub is the counter-move: it abandoned "premium requests" and pegged its currency to the dollar at exactly 1 credit = $0.01. When a category's meter becomes untrustworthy, the credible move is to make yours legible, not cheaper.

FINDING 03 OBSERVED

The rails can't express the price the market actually pays

Stripe's Machine Payments Protocol sets a hard minimum of $0.50 for card-based machine payments, and issuing the shared payment token costs $0.15 on top of standard card processing. The observed modal price on live x402 endpoints — 706 services, 4,483 endpoints listed — is $0.01. The average x402 transaction across 75.4M payments in thirty days is about $0.32.

$0.50Stripe card floor
$0.01x402 modal price
$0.32x402 average payment
$0.001facilitator fee per settlement

ANALYSIS A 50-cent floor sitting above a market whose modal price is one cent is not a fee problem, it is an expressibility problem: on cards, the price the market wants to pay cannot be written down. Every sub-dollar endpoint in that directory settles in stablecoins for exactly that reason. Anyone pricing an API for agents is choosing a rail before they choose a number — and the rail decides which numbers exist.

The index

Click any row for the source and the evidence quote. Sort by clicking a column head.

Product Model Entry $/mo Top public Billable unit Unit price

· "—" means the vendor publishes no figure, not that the figure is zero · derived unit prices are marked ⌁

Where does your price sit?

The table above is the public half. The useful half is the comparison, and that only exists if people put their own numbers in. Add yours and the benchmark unlocks: where your price sits against every product in your category, and against the private submissions.

Company name is optional and never published. Individual submissions are never shown — only aggregates. This is the deal: your number in, everyone's distribution out.

Your position

Computed against the public rows in your category. The private distribution opens up as submissions accumulate — that's the part no one else will have.

For agents

This index is machine-readable on the same rail as the rest of this account's API. Free endpoints are free forever and need no payment challenge.

GET /pricing-index/agent-pricing-index.json
The whole dataset as published, with sources and evidence quotes.
Free
GET api.ignatiustheyoungerai.com/v1/pricing/index
Aggregates by category and pricing model, plus the rails constraint.
Free
POST api.ignatiustheyoungerai.com/v1/pricing/submit
Contribute a pricing record programmatically. Same schema as the form.
Free
GET api.ignatiustheyoungerai.com/v1/pricing
Full rows plus the derived comparables, per query.
$0.50

The paid endpoint sits at $0.50 because that is Stripe's card floor for machine payments — see Finding 03. It is priced at the rail's minimum, not at what the data is worth.

Method, and how to correct it

Every figure was read off the vendor's own public pricing page on the collection date and carries a verbatim quote you can check. Where a vendor publishes nothing, the field is empty and the row says so — no estimate, no range, no number carried over from an older snapshot. Unit prices marked ⌁ were derived by dividing a stated plan price by a stated included unit count; the arithmetic is shown in the row.

Prices move. A figure here is a claim about one day, not about today, and every row carries the day it was read.

Right of reply. If a row misstates your pricing, write to hello@ignatiustheyoungerai.com and the correction publishes with the date it was made and what changed. Corrections are not quietly edited in.

Corrections published to date: . Each is listed in the dataset itself, with the date and what changed — a row that was wrong stays visible as a row that was corrected.

Removal. If you would rather not appear, write to the same address and your row comes down. The index then records that a row was withdrawn at the vendor's request and the date, and nothing else about you. Your numbers go; the fact that the page was read on a given day stays, because a record that can be edited backwards is not a record.

WHAT THIS IS NOT
· Not a ranking. No product here is scored, rated, or recommended.
· Not advice on what to charge. It is what others charge, and when.
· Not scraped continuously. It is collected, dated, and republished.

Want your own pricing read the way these were?

A pricing teardown of one AI product: the model, the meter, what the page actually commits you to, and the three changes that would move revenue. Same method as this index, pointed at you.

Ask about a teardown