AI credits explained: five vendors, four units, one published exchange rate
Every AI tool now sells you an allowance. Almost none of them tell you what one unit of that allowance buys, which means the advertised price is only half the contract.
- Four different invented units across five vendors. AI credits, credits, priority credits and Fast Tokens. None of them convert.
- Only Plus AI publishes a rate. It states tasks typically use 40 to 170 credits, which makes 1,500 credits roughly 9 to 37 tasks a month.
- The other four publish an allowance and no rate, so their numbers cannot be turned into work before you pay.
- Beautiful.ai refuses the whole mechanism and advertises unlimited AI generation with no metered unit at all.
- An allowance can be revalued without changing the price. That is the buyer-relevant property of credit pricing, and it is why the unit matters more than the number.
The finding, in one table
Five vendors across two categories, read from their own pricing pages in August 2026. The last column is the one nobody publishes.
| Vendor | Calls its unit | Allowance published | Rate published |
|---|---|---|---|
| Plus AI | credits | 1,500 · 3,000 / mo | Yes: 40 to 170 per task |
| Gamma | AI credits | 400 · 1,000 · 4,000 · 20,000 | No |
| Ideogram | priority credits | 1,000 · 1,500 · 3,500 / mo | No |
| Leonardo | Fast Tokens | 150 / day · 8,500 to 60,000 / mo | No |
| Beautiful.ai | no unit | "Unlimited AI generation" | Not applicable |
Read from gamma.app, plusdocs.com, beautiful.ai, ideogram.ai and leonardo.ai pricing pages, 25 August 2026. Rates and allowances exclude tax.
What an AI credit actually is
A per-vendor unit of usage, defined by the vendor that issues it, redeemable only inside that vendor's product.
A credit is not a currency and it is not a standard. It is closer to an arcade token: minted by one operator, worth whatever that operator says today, worthless everywhere else. The word appearing on three different pricing pages does not mean those three vendors are selling the same thing, and the fact that a fourth calls its version a Fast Token instead is the honest signal that no shared definition exists.
A published rate
÷ 40 to 170 per task
= 9 to 37 tasks
No published rate
÷ not published
= ?
That FAQ entry is the most useful thing on the page, because it is a vendor conceding that its own unit is not self-explanatory to the people buying it.
Why the unit matters more than the number
Because of what each one lets a vendor change quietly.
A price is public and sticky. Raising it is visible to every existing customer, it generates coverage, and it invites cancellations. An allowance has neither property. If a vendor decides that an operation which used to cost 40 credits now costs 90, your effective cost has more than doubled and the number on the pricing page has not moved. Nobody publishes a changelog for that.
Why vendors do this, stated fairly
There is a real problem underneath it. A short text prompt and a long high-resolution video generation can differ in underlying compute cost by orders of magnitude, and a flat per-request price would either overcharge the first or lose money on the second. A credit is a reasonable abstraction over genuinely variable cost, and every vendor here is solving that problem rather than inventing a trick.
The criticism is narrower and it stands: solving that problem does not require withholding the conversion. Plus AI meters in credits and still publishes a range, and publishing a range cost it nothing. The other four could publish "a standard generation costs roughly N" tomorrow. That they have not is a choice about disclosure, not a consequence of the pricing model.
Rollover, banking and what happens to unused credits
Mostly unstated. Leonardo is the clearest of the five, and what it publishes is a ceiling rather than a promise.
| Leonardo plan | Monthly allowance | Token Bank ceiling | Ratio |
|---|---|---|---|
| Free | 150 Fast Tokens / day | 150 | daily, not monthly |
| Essential ($12) | 8,500 / mo | 25,500 | 3 months |
| Premium ($30) | 25,000 / mo | 75,000 | 3 months |
| Ultimate ($60) | 60,000 / mo | 180,000 | 3 months |
The ratio is the interesting part: the ceiling is exactly three months of the plan's own allowance at every tier. That is a deliberate, consistent policy rather than an arbitrary number, and it is the single clearest disclosure any of these five vendors makes about its own unit. It also tells you what banking is for. You can carry a slow month into a busy one, and you cannot hoard a year.
What "unlimited" means when it appears
Three of the five advertise unlimited something. Only one of them means unlimited output.
Unlimited relaxed generation on selected named models. Its own footnote states concurrency may be slowed from time to time based on demand.
Unlimited slow credits on every paid tier, with priority credits metered separately. The metered unit buys queue position, not access.
Advertises unlimited AI generation with no credit system published. The most legible of the five, and also the most expensive per seat.
What to check before buying anything metered
1. What does one unit buy? If the pricing page does not say, you are buying an allowance you cannot value. Ask support in writing before subscribing annually.
2. Is there more than one unit? Ideogram meters priority credits and slow credits separately; Leonardo meters Fast Tokens and offers relaxed generation alongside them. A single headline number can hide a two-tier system.
3. What happens to what you do not use? A published bank ceiling, like Leonardo's, is a real answer. Silence usually means it expires.
4. Does the allowance reset daily or monthly? Leonardo's free tier is 150 tokens per day, which behaves nothing like a monthly pool of the same total.
5. Is anything free actually conditional? Ideogram's free slow credits are described as available "weekly for eligible accounts," and eligibility is not defined on the pricing page.
How this was checked, and when it expires
Every figure on this page was read from the vendor's own published pricing page on that date and excludes tax. No figure here is estimated, modelled or carried over from an earlier version of this article.
Allowances change more quietly than prices do, which is the entire argument of this page and therefore also its expiry condition. If a number here no longer matches the vendor's own page, the vendor's page is right and this one is stale. Tell us and we will re-read it.