Every AI product meters something, and no two vendors meter the same thing. The word “credit” appears on most pricing pages and means something different on each. This is a guide to reading those pages, so you can compare two plans that look similar and price the work you actually do.
The four meters you will meet
Tokens measure text. One token is roughly three quarters of an English word, and both what you send and what comes back are counted. This is the unit of language model APIs, and it is the only one with a common definition across vendors.
Credits measure a completed action: an image, a second of video, a translated document. Each vendor sets what a credit buys, which is why the number on a pricing page tells you nothing on its own.
Time measures compute. Midjourney sells fast GPU hours, so a complex generation costs more of your allowance than a simple one, and its $10 plan buys 3.3 hours a month.
Executions and tasks measure automation runs. The distinction matters more than it sounds: Zapier counts every step in a workflow, while n8n counts one whole run whatever it contains. A forty-node workflow costs forty times more on one model and the same on the other.
Why the same word means different things
Ideogram publishes its rates plainly: on the 4.0 model at 2K, one credit buys one image in Turbo, two in Balanced and three in Quality. Its $20 plan includes 1,000 priority credits, so the honest figure is between 333 and 1,000 images depending on how you work.
Adobe Firefly uses the same word for something else entirely. Standard image features run unlimited on every paid plan and cost no credits at all; credits are only consumed by premium work such as video generation, translation and partner models. Its 2,000 credits and Ideogram’s 1,000 are not comparable quantities.
Make prices its plans “for 10,000 credits a month”, where a credit is one module execution. A five-step scenario running a hundred times spends around five hundred. The subscription price is a floor, and the credit tier is the actual lever.
How to price one finished result
The number to compute is the cost of one thing you would ship, including the attempts it took.
Take image generation. If a usable image takes four attempts at Quality on Ideogram, that is twelve credits, so a 1,000-credit plan yields about eighty finished images a month rather than a thousand generations. Do that arithmetic before comparing two plans.
The same applies to video, where the gap is widest. A plan advertising forty clips a month may deliver eight seconds of usable footage, because a shot commonly takes several generations to land.
The three questions that settle a pricing page
Does the meter reset, roll over or expire? Suno credits do not carry from month to month. Otter minutes have no rollover. Leonardo runs a token bank that does carry them forward, up to a cap. This changes the value of a plan by a lot more than a few dollars of price difference.
What happens when it runs out? Zapier switches to pay-as-you-go so the automation keeps running unless you turn that off. Others pause until the next cycle. Neither is wrong, but only one of them can produce a surprise invoice.
Which actions are free? On several products the common action costs nothing and only the premium one is metered. If your work is mostly the free action, a smaller plan may be enough.
What this means for choosing
Ignore the headline credit count. Two numbers matter: the cost of one finished result at the quality you ship, and the behaviour at the ceiling.
Every listing in this directory states the meter, the allowance and the date its pricing page was checked, precisely because the pricing pages themselves rarely make this comparison possible. If you want to see the range, the pay-as-you-go tools are the ones where consumption is the whole model rather than a limit on a subscription.
