AI credits explained
What a credit is, how credits are calculated from the actual cost of each AI call, how reservations, expiry and refunds work, and where to see every debit.
Updated October 14, 2026
On this page
Credits are not tokens
Credits are CVRLanding’s unit for AI work: generating a page, editing it in chat, analyzing a URL, drafting a variant or diagnosing a funnel. AI providers bill in tokens, and each model prices input, output and cached tokens differently. Credits turn that actual cost into one balance you can plan with.
1,000,000 credits is an amount of AI work, not 1,000,000 model tokens. The same request can cost more on one model than on another.
How credits are calculated
Each AI call is charged from its actual provider cost, converted at a published cost ratio:
retail value =
actual provider cost ÷ cost ratio
credits = ceil(
retail value × credits per USD
)By default the cost ratio is 30% and there are 100,000 credits per US dollar, so $1.00 of provider cost is 333,334 credits. The calculation uses exact whole numbers and rounds up once, at the end, so a debit is never more than one credit above the exact value.
The ratio is not a surcharge or a token discount. It is the share of the credit value that goes to the AI provider: at 30%, the provider cost is 30% of the value of the credits debited.
Worked examples
Credits charged for three provider costs, at the default ratio of 30% and at 20%. These are example costs to show the arithmetic, not quotes: what a real request costs depends on the model and on how much text goes in and out.
$0.01
- At 30%
3,334
- At 20%
5,000
$0.25
- At 30%
83,334
- At 20%
125,000
$1.00
- At 30%
333,334
- At 20%
500,000
| Provider cost | At 30% | At 20% |
|---|---|---|
$0.01 | 3,334 | 5,000 |
$0.25 | 83,334 | 125,000 |
$1.00 | 333,334 | 500,000 |
A lower ratio means more credits for the same provider cost, because each credit stands for less of it.
The cost ratio
The default is 30%. CVRLanding’s billing admin can set it between 5% and 90%.
Each change is a new version with an effective date. Requests from that date use the new ratio.
Every debit records the ratio version, the provider cost, the model pricing version and the cost basis: reported by the provider, calculated from the model’s price table, or estimated.
Changing the ratio never recalculates past debits.
Credits in each plan
Each plan grants credits for its period. The last column is the most provider cost the grant can cover at the default ratio if every credit is spent: credits ÷ credits per US dollar × ratio.
Trial
- Credits
20,000 one-time
- Max AI cost
$0.06
Starter
- Credits
1,000,000 per month
- Max AI cost
$3.00
Growth
- Credits
2,900,000 per month
- Max AI cost
$8.70
Pro
- Credits
7,900,000 per month
- Max AI cost
$23.70
Agency
- Credits
19,900,000 per month
- Max AI cost
$59.70
| Plan | Credits | Max AI cost |
|---|---|---|
Trial | 20,000 one-time | $0.06 |
Starter | 1,000,000 per month | $3.00 |
Growth | 2,900,000 per month | $8.70 |
Pro | 7,900,000 per month | $23.70 |
Agency | 19,900,000 per month | $59.70 |
Prices and current plan details are on the pricing page.
Reserve, execute, settle
Reserve. Before a request runs, the worst-case estimate is reserved from your balance. For larger tasks you see the estimate first. If your balance can’t cover the reservation, the request doesn’t start.
Execute. The call to the AI provider runs under a job ID, with a limited number of retries.
Settle. When the provider reports actual usage, its cost is converted with the formula and debited, and the rest of the reservation is released.
If the actual cost would exceed the reservation, the job stops at the authorized maximum and returns a partial result with a clear status. Your balance never goes negative.
If a provider doesn’t report usage, the job is settled from a trusted estimate, marked Estimated and reconciled later against the provider’s reports. An estimate is never shown as confirmed, and any correction is added as a new ledger entry rather than an edit.
Failed jobs and retries
The same attempt is never debited twice, even if a queue delivers it again: every debit carries a unique key. Separate attempts are different. If a provider bills for a failed attempt, that cost is real, and it is recorded as it happened.
Whether credits for a failed job are refunded follows CVRLanding’s refund policy for failed jobs. A refund appears as its own ledger entry. See billing and refunds.
When credits expire
Monthly grants expire at the end of the billing period, so unused monthly credits don’t carry over.
Trial credits are a one-time grant and expire when the trial ends.
Top-ups, where offered, are separate grants with their own validity.
Spending always draws from the grant that expires first.
When credits run out
New AI operations pause until credits are available again: when your next billing period starts, or after you upgrade or buy a top-up where one is offered.
Everything else stays yours. You can still open your projects, view and export your data, and manage billing. Published pages are not deleted because credits reached zero: whether a page stays online depends on your subscription and plan, not on your credit balance. Credits only pay for AI work, so visitors filling in forms on your pages don’t use them.
Where to see your credits
Open Credits & billing. The balance shows credits allocated, reserved, used and remaining, where remaining is allocated minus used minus reserved.
The ledger lists every grant, reservation, settlement, release, refund, expiry and adjustment. Each AI debit shows the job, the model, the provider’s token counts, the provider cost, marked as actual or estimated, and the ratio version used.
Plans and prices are on the pricing page. Invoices, payment methods and cancellation are handled in the Paddle Customer Portal, which you open from Credits & billing.