How AI tool pricing actually works: per-seat, usage, and the meters in between

AI tools bill seven different ways and most combine a few at once. Here is a plain-language guide to the pricing models, what each one hides, and how to tell what you will really pay.

Shop for AI tools for a week and you will notice that no two pricing pages agree on anything. One charges per seat. The next charges per “resolution.” A third quotes per minute, or per credit, or refuses to quote you anything until you book a call. You are not missing something. The category genuinely has not settled on how to charge, which makes comparing options harder than it should be. Here is the map.

The seven ways AI tools bill you

Per seat. The familiar one: a monthly price per user. Predictable, easy to budget, and the default most teams reach for. The catch is that the seat rarely covers the actual AI work, which gets metered separately underneath.

Usage, or per token. You pay for what the model processes. Common with developer tools and anything that hands your work to an underlying LLM. Honest, but you cannot really forecast it until you have run live volume.

Per credit. A buffer in the middle. You buy credits and each action spends some. The problem is that one credit means different amounts of work in different tools, so credit prices across two products tell you nothing on their own.

Per resolution. Big in support. You pay only when the AI fully handles a ticket, often fifty cents to a dollar each. Lovely when it works, because cost tracks value, but it scales with success, so the bill climbs exactly when the tool is earning its keep.

Per minute. The voice-agent model. You pay for call time, and the real rate quietly stacks the platform fee, the model, and telephony on top of each other.

Per task. How automation platforms count. Every step or run costs something, so a multi-step workflow adds up faster than anyone expects.

Quote-only. No public price at all. About one in five of the tools we track sit here, which usually means enterprise contracts and onboarding fees rather than a signup.

Most tools use more than one

This is the part that catches people. More than half the tools we track combine two or more of these at once. A per-seat plan with usage credits on top. A platform fee plus per-minute charges. A flat tier that includes some allowance, then meters everything past it.

So the plan you sign up for is the floor, not the total. When a tool shows you one clean monthly number, the real question is what sits under it, and how fast that second meter moves once you actually use the thing.

How to read a pricing page in two minutes

Find the second number. Scroll past the headline plan to the line that reads “then $X per Y.” That Y, a resolution, a minute, a credit, a thousand tasks, is where your money goes.

Then map it to your volume. The same tool can be cheap at low volume and brutal at high volume, or the other way round. Use your own numbers, not the vendor’s flattering example.

And split predictable from variable. Seats and platform fees in one column, everything metered in another. The second column is the one that surprises people, so it is the one to keep an eye on.

The short version

There is no standard unit, so stop comparing headline prices as if there were. Find the meter, model it against your real usage, and watch the variable column. Do that and AI pricing turns from a mystery into a spreadsheet.

We keep the sourced, dated breakdown current in the AI Agent Pricing Index, with the real entry cost and billing model for every tool we track.