Updated August 2026·12 min read

How to read an AI pricing page: six patterns, fourteen vendors

Over one week we priced fourteen AI vendors across six categories, reading every figure from the vendor's own page. The same six patterns turned up almost every time, and not one of them is the number printed in the largest font.

Last reviewed: August 2026 Next review: November 2026
Bottom line up front

The six checks, in order

Ordered by how much money each one tends to be worth, not by how obvious it is. Every example below was read from a vendor's own pricing page in August 2026 and is linked to the full working.

What is actually being counted?

Before any comparison, identify the unit. In one week we recorded ten: credits, AI credits, priority credits, Fast Tokens, characters, minutes of audio, minutes of video, videos per month, orders per month, recruiter seats, company headcount and annual revenue band. Two figures are only comparable if they price the same noun.

The sharpest case: Manatal bills per recruiter seat and Workable bills on your company's total employee count. For two recruiters at a fifteen-person company that is $30 a month against $299 for the same work.
Which meter do you hit first?

Most plans carry more than one limit, presented with equal visual weight, and only one of them will bind you. Work out which, then price only that. This is the check that most often changes which plan you should buy.

Sellerboard Standard allows 3,000 orders a month and 150 follow-up emails, five percent coverage; the next tier up costs $8 more and raises emails forty-fold to 6,000. Submagic Starter allows fifteen videos at a maximum of two minutes each, and the two minutes is what stops a podcast clip.
Does the vendor publish what one unit buys?

An allowance without a conversion rate cannot be turned into work before you pay. Most vendors publish the first and withhold the second, so a credit total on its own tells you nothing.

Of five credit-metering vendors, only Plus AI published a rate on its pricing page (40 to 170 credits per task). VEED publishes one that is internally consistent at 4 credits per AI video across all three tiers. ElevenLabs is the most complete: seven published rates, from 1 credit per character to 10,000 per minute of studio dubbing. Gamma, Ideogram and Leonardo publish allowances and no rate.
What is the free tier actually withholding?

Frequently not a quantity. The paid feature is often the removal of a restriction the vendor chose to add, which never appears in a comparison of allowances and decides whether the free tier is usable at all.

Leonardo makes free creations public; privacy starts at $12. ElevenLabs withholds the commercial licence until $6. OpusClip stops free clips being exportable after three days, and even its $15 tier caps retention at thirty. Ideogram's free slow credits are available "weekly for eligible accounts", with eligibility undefined.
What does "unlimited" have attached to it?

Read the rest of the sentence and then the footnote. In this market the word almost always means unlimited-slow, unlimited-on-a-subset, or unlimited-until-we-say-otherwise.

Leonardo: unlimited relaxed generation on selected models, and its own footnote reserves the right to slow concurrency. Ideogram: unlimited slow credits, with priority credits metered separately. Workable: unlimited active jobs, subject to fair usage limits that carry no number. Beautiful.ai was the only vendor surveyed advertising unlimited AI generation with no metered unit at all.
Is the annual discount real, and is the cheap plan cheap?

Two multiplications settle it. Some advertised savings are exact, some attach to something other than the billing period, and occasionally the entry plan is the worst purchase on the page.

Helium 10 is exact: $129 minus $99 across twelve months is precisely the $360 advertised. Workable shows $299/month and $3,588/year, which is exactly twelve times the monthly rate; its 20 percent saving attaches to bundling recruiting with HR. OpusClip Pro billed annually is $14.50 a month, cheaper than its own $15 Starter plan, which is offered on monthly billing only, with 3,600 credits against 150.

Ten meters, six categories

The full list of what these vendors count, which is the single most useful thing to know before opening a comparison table.

CategoryVendorMetersPublishes a rate?
PresentationsGammaAI creditsNo
Plus AIcreditsYes, 40 to 170 per task
Beautiful.aino unitNot applicable
Image generationIdeogrampriority creditsNo
LeonardoFast TokensNo
Voice and audioElevenLabscredits, one shared poolYes, seven rates
Short-form videoOpusClipcreditsNo
Submagicvideos, minutes and AI creditsNo
VEEDcreditsYes, 4 per AI video
PodcastingPodsqueezeminutes of podcast timeUnit is self-explanatory
RecruitingManatalrecruiter seats, jobs per accountUnit is self-explanatory
Workablecompany headcountUnit is self-explanatory
Amazon seller toolsHelium 10annual revenue bandUnit is self-explanatory
Sellerboardorders per month, emails per monthUnit is self-explanatory
Notice which column is full and which is empty. Every vendor that meters in a unit you can picture, minutes or orders or seats, needs no conversion rate because the unit explains itself. Every vendor that invented a unit needs one, and most of them do not provide it. The invented unit and the missing rate are the same decision seen twice.

The one number that keeps repeating

Of everything measured this week, the strangest finding is a convergence nobody advertises.

Leonardoimages, Fast Tokens3x
ElevenLabsvoice, credits3x
Podsqueezepodcasting, minutes3x

Maximum accumulated allowance, as a multiple of the monthly quota. Three vendors, three categories, three different units.

Leonardo's is the one that proves it deliberate: its Token Bank is 25,500 on a plan granting 8,500 a month, 75,000 on 25,000, and 180,000 on 60,000. Three tiers, three different numbers, every one exactly three months of that tier's own allowance. All three vendors condition it on staying subscribed, and none extends it to a free plan.

The honest scope of that claim. It is three vendors out of eight we attempted. Two more, Descript and Speechify, were fully readable and publish no rollover terms at all; three others returned blocked or JavaScript-only responses and are counted neither way. So the finding is narrow and checkable: of the vendors that publish a rollover cap, all of them so far chose 3x. A vendor publishing 2x or 6x would break it, and we would like to know.

Why the price is the half that does not move

One structural observation that explains most of the patterns above.

A published price is sticky. Raising it is visible to every existing customer, it generates coverage, and it prompts cancellations. An allowance has neither property. If an operation that used to cost 40 credits starts costing 90, your effective cost has more than doubled and the number on the pricing page has not moved by a cent. Nobody issues a changelog for that, and no comparison site tracks it.

This is not an accusation. Metering compute genuinely is hard, costs really do vary by orders of magnitude between a short prompt and a long video render, and a credit is a reasonable abstraction over that. But it has a side effect that falls entirely on the buyer, and it is why every check on this page is aimed at the allowance rather than the price.

The practical rule. The advertised price tells you what you pay. The undisclosed rate tells you what you get. Watching only the first one means watching the half that is designed to stay still.

How this was compiled

Fourteen vendors · six categories · August 2026

Every figure cited here was read from the vendor's own published pricing page between 25 and 26 August 2026 and excludes tax. Nothing is estimated, modelled, or taken from a third-party comparison. Where a vendor's page could not be read, whether through bot protection or JavaScript-only rendering, no figure is published and the vendor is named as unread rather than counted as absent.

Each pattern above links to the full working, including the plan tables, the arithmetic and the vendors that could not be priced.

This page ages faster than most. Allowances, caps and rates move more quietly than prices do, which is the entire argument and therefore also the expiry condition. If a figure here no longer matches the vendor's own page, the vendor's page is right and this one is stale.

Frequently asked questions

What should you check first on an AI pricing page?
What the vendor is counting. Across fourteen vendors the meter was variously credits, Fast Tokens, characters, minutes of audio, minutes of video, videos per month, orders per month, recruiter seats, company headcount and annual revenue band. Two prices are only comparable if they price the same thing, and in this market they usually do not.
What does "unlimited" mean on an AI pricing page?
Almost always unlimited slow, unlimited on a subset, or unlimited subject to a fair-usage clause with no published number. Leonardo restricts unlimited relaxed generation to named models and reserves the right to slow it; Ideogram gives unlimited slow credits while metering priority credits; Workable qualifies unlimited active jobs as subject to fair usage. Beautiful.ai was the only one advertising unlimited AI generation with no metered unit at all.
Do AI credits expire or roll over?
Where a vendor publishes a policy, the cap has so far always been three times the monthly allowance. Leonardo's Token Bank is exactly three months of each tier's allowance, ElevenLabs caps a balance at 3x the monthly quota, and Podsqueeze allows up to 3x monthly minutes. All three require an active paid subscription and none applies to a free plan. Where nothing is stated, assume it expires.
Why is a free AI tier usually not usable for real work?
Because the limit is often not a quantity. Leonardo makes free creations public, so privacy begins at $12. ElevenLabs withholds the commercial licence until $6. OpusClip stops free clips being exportable after three days. In each case the paid feature is the removal of a restriction rather than more compute, and none of it appears in a credit comparison.
Is the annual discount on AI tools always real?
No, and two multiplications settle it. Helium 10's advertised savings are exact. Workable displays $299 a month and $3,588 a year, which is exactly twelve times the monthly rate, with its 20 percent saving attached to bundling rather than to billing period. OpusClip's Pro plan billed annually is cheaper per month than its own Starter plan, which has no annual option.
What is the fastest way to compare two AI tools honestly?
Work out which meter you personally hit first, then price only that. Sellerboard's entry plan allows 3,000 orders but 150 emails, so anyone sending review requests is bound at a twentieth of the headline. Submagic caps video length at two minutes, which binds long before its fifteen-video allowance. The binding constraint is rarely the number in the largest font.

The full working

Save
Dashboard

From our network

Best AI Tools for Amazon Sellers - bagengine.comBest AI Courses 2026 - edubracket.comBest Accounting Software for Online Sellers - ceocult.com