Usage-based vs seat-based AI pricing: which costs you less
Seat pricing charges per person regardless of use; usage pricing charges per token, credit or image with no floor. Bursty and occasional work is cheaper on usage; daily heavy work is usually cheaper on a seat. Most overspending comes from being on the wrong model, not the wrong vendor.
By the AI Alt Finder team
AI tools price in two shapes and a hybrid. Seats charge per person per month whether or not that person logs in. Usage charges per unit consumed, whether that unit is a token, a credit, an image or a request. The hybrid, now common, sells a seat that includes a usage allowance and bills overage beyond it.
Choosing the wrong shape costs more than choosing the wrong vendor within a shape, and it is a much easier mistake to fix.
When does seat pricing win?
When usage is high and steady. A flat subscription is effectively an unlimited-use insurance policy: you pay a known amount and stop thinking about consumption, which also removes the quiet productivity tax of hesitating before every expensive action.
Two concrete examples of the shape, both checked on 28 July 2026: Anthropic lists Claude Code as included from its Pro plan at $20 per month billed monthly, and GitHub prices Copilot Pro at $10 per user per month. Neither meters you per request at those tiers, which is the whole point.
When does usage pricing win?
When work is bursty, seasonal or occasional. A metered key costs nothing in a quiet month, which is exactly when a seat is pure waste. It also scales to zero when someone goes on holiday or a project ends, and nobody has to remember to remove a seat.
The second case for metered is multi-provider freedom. Open-source clients licensed Apache-2.0, such as Aider, Cline, Continue and the vendor CLIs from OpenAI and Google, all run against whichever key you supply, so you can move providers without changing tools.
Where do credits fit?
- Credits are usage pricing sold as a prepaid block, which shifts the risk of estimating demand from the vendor to you.
- Watch expiry: a monthly credit grant that does not roll over turns unused capacity into a loss every month.
- Watch variable burn: where a credit buys a task rather than a fixed unit, one complex task can consume several, so the headline allowance is not a count of actions.
- Watch what credits do not cover. Export, custom domains, backend features and commercial licences are commonly plan features rather than credit purchases.
How do you work out which one you are on the wrong side of?
Take last month. If you are on a seat, estimate what the same work would have cost metered at published per-unit prices. If you are metered, compare your bill to the seat price of the nearest equivalent subscription. Do it for three months, not one, because a single month hides seasonality.
A gap under about twenty percent is not worth a migration, given the hours a switch costs. A gap of two or three times means the shape is wrong, and that is worth acting on.
The trap in hybrid pricing
Seat-plus-allowance plans are the hardest to forecast, because your bill is stable until it is not. The allowance makes light months look like a flat subscription, then a heavy month arrives as overage nobody budgeted for.
If you buy a hybrid plan, find the overage rate before you sign and calculate what a bad month costs. If the vendor does not publish that rate clearly, treat the omission as information.