Methodology

What the multiple is

For every product we state a usage assumption — a specific, boring description of a month's work. We price that usage against a dated rate card of published API prices and divide the subscription price by the result. That ratio is the markup multiple.

It is an estimate of the gap between what you pay and what a comparable month of inference would cost at list price. It is not a claim about a vendor's costs, margins, or profits. We do not know those. Nobody outside these companies does.

Three things push a vendor's true cost below our estimate: volume discounts, in-house serving, and smaller models than we assume. That makes every multiple here a floor. We would rather understate a markup than overstate one.

Which price we use. Every entry uses the month-to-month price — what you pay without committing to a year. Most vendors advertise a lower annual rate, and where they do, it is recorded in that entry's billing note, usually along with what it does to the multiple.

This rule exists because the directory ranks products against each other. Quoting one product's annual rate against another's monthly rate produces a table that looks comparable and is not: a product can halve its apparent markup purely by which of its own advertised prices we happen to pick. If you are an annual subscriber, your multiple is lower than the one shown here, often by a lot — the entry says by how much.

Where a vendor prices differently by region, we use the US list price and say so, because it is what most readers of this site are shown. That is a choice, not a fact about the product, and it is recorded on the entry rather than hidden in it.

Why the multiple never decides the verdict

A high multiple is a question, not an answer. The question is: what else is this price buying? Sometimes the honest answer is "a great deal" and sometimes it is "an interface".

The clearest illustration is inside this directory right now. Midjourney Standard carries an estimated 36× markup and is rated ●FAIR. Monica Pro carries 2.6× — far less — and is rated ▣WRAPPER. The difference is not arithmetic. It is that one sells a model trained in-house with no API to wrap, and the other runs the pipeline described in every retrieval tutorial published since 2023.

A 40× markup on a compliance-bound medical scribe can be fair. A 4× markup on a prompt in a text box can be a wrapper. If you find an entry where the verdict looks like it was read off the number, that is a mistake and we want the correction.

The three verdicts

On the word "wrapper". It describes an architecture: a user interface and a prompt over a third-party model. Building one is legal, common, frequently sensible, and sometimes exactly what a customer wants to buy. It is not an accusation of dishonesty, and we do not intend it as one. If a verdict on this site reads as an allegation of misconduct, the wording is wrong and we will fix it.

The rate card

Entries are priced against archetypes — "mid-tier frontier text model" — rather than named products. Naming one vendor's model would make every estimate on this site wrong the day that vendor repriced. Each archetype is set at a representative published rate and links to a real rate card you can check.

Rate card updated 11 Aug 2026.

Costs are computed from this table at build time. No entry stores a cost or a multiple of its own, so when a rate changes here every affected figure on the site changes with it in one commit — and none of them can silently disagree with the arithmetic.

How a price gets published

Three states, and only the last one reaches production:

• unverified — a draft. Nobody has confirmed it against the vendor's own page. Shown with a warning, and the deploy gate refuses to pass.
• auto-observed — the price watcher found this exact price on the vendor's live pricing page on a stated date. Machine evidence, no human judgement. Still not publishable on its own.
• verified — a named person checked it on a stated date and their name is in the commit.

A weekly watcher re-fetches every cited pricing page and compares it to the last snapshot. If a page moves, or if the recorded price stops appearing on it, the entry is flagged and can be automatically demoted back to unverified — which fails the deploy until a human looks. Entries older than 90 days are marked stale on their own page.

The watcher detects change; it does not read prices. Pricing pages are rendered client-side, split-tested and localised, and any tool claiming to reliably read a price off one is overselling itself. Noticing that a page moved is robust, and it is enough — a person confirms the number.

What this site does not claim

• Not a claim about any vendor's actual costs, margins, or profitability.
• Not a claim that any price is unfair, deceptive, illegal, or worth what you would pay elsewhere.
• Not a claim that any company has done anything wrong.
• Not financial, legal, or purchasing advice.
• Not a measurement. Every cost figure is an estimate, computed from stated assumptions, and labelled as one.
• Not affiliated with, endorsed by, or sponsored by any product listed.

Product names, logos, and trademarks belong to their owners and are used here only to identify the products under discussion — the ordinary descriptive use that writing about products requires.

Verdicts are opinion, formed on the stated methodology and published alongside the figures and sources they rest on, so you can disagree with the reasoning rather than take our word for it. Where we state facts — a price, a billing term — we date them and link the source.

Corrections

If a number here is wrong, we want to know, and we would rather hear it from you than not hear it. Two routes, both fine:

• Open an issue or pull request against the entry's JSON file in the repository, where every figure's full history is public.
• Email office@isitjustawrapper.com. Vendor mail gets the same treatment as anyone else's, and you do not need an account anywhere to use it.

What we commit to. Anything factual — a wrong price, a wrong tier, a misstated billing term — is corrected on sight, without argument, and logged. Disagreements about a verdict get a reply and, where the argument lands, a revised entry; where it does not, the vendor's position is recorded on the entry so readers see both. Every change appears on the corrections log with a date and a description, including the ones we caught ourselves.

That log is the point. A reference that never publishes its mistakes is not more accurate than one that does — it is just harder to check. 30 entries are currently listed.

Money

The site sells 20 sponsor slots. Sponsors are labelled as sponsors everywhere they appear, their links carryrel="sponsored", and impression and click totals are published on stats.

Sponsorship has no effect on a verdict, and cannot buy one. A sponsor may be listed in the directory and may receive any verdict, including ▣WRAPPER. If that ever appears to have been violated, the sponsorship is the thing that goes, not the entry.