Claude

What Claude actually meters

Claude meters a seat, and the cap on that seat is the real price. Two windows govern it: a rolling five-hour session on every plan, and a weekly cap on top for paid plans. The weekly window resets at a fixed hour assigned to your account, not on your billing date, and Settings shows two weekly bars, one for Opus and one for everything else. A per-model sublimit exists, then, even where the plan pages claim the weekly limit covers all models.

Anthropic is explicit that there's no fixed message count. Message length, attachment size, conversation length, tool use, model choice and effort level all move it, and cached project knowledge cuts it. That's an honest description of a token meter, just not a number.

The quantities are the gap. Pro is "at least 5x" Free, Max is 5x or 20x Pro, Team Standard is 1.25x Pro and Premium 6.25x. Every rung is relative and the base rung is undocumented. Fable models carry their own sublimit: Max and Team Premium seats can spend up to 50% of a weekly limit on them, while Pro and Team Standard run them on usage credits from the first message. Claude Code isn't a separate budget.

How credits work

Claude's usage credits are prepaid dollars, not a synthetic unit, and they bill at standard API rates. There's no conversion to work out, because the API rate card is the price list.

They sit behind a toggle. Pro and Max users enable them in Settings, prepay a balance and either set a monthly spend cap or leave it unlimited. Auto-reload refills below a threshold you set, and a $2,000 daily redemption limit applies.

Credits generally don't expire, with one exception: in jurisdictions such as Japan they expire six months after purchase, effective 10 September 2026. On Team plans an owner buys a shared balance and sets organisation, seat-tier and per-user caps. Self-serve Enterprise pools credits, so one heavy user drains everyone's balance. Sales-assisted Enterprise invoices monthly in arrears instead.

What happens when you hit the limit

Claude blocks and tells you when the window resets. It doesn't throttle and it doesn't charge you by default.

From there you wait, upgrade, or turn on usage credits. Free users only get the first two. Where credits are on, the switch is explicit: a notification fires at the limit and you confirm before consumption billing starts.

Spend controls stack across organisation, seat tier and user, and the lower limit wins. Hitting a user cap pauses that person until the next billing period. Anthropic also documents the leak: it checks your limit before a request and counts tokens after, so one request can carry you past the cap before the next blocks. Few vendors write that down.