Plans and usage
How a subscription's two caps work, what happens when you reach one, and how metered usage differs from a plan.
Two caps, not one
Every plan carries two independent limits, and they do different jobs:
- The window cap is a rolling short-term ceiling. It bounds how hard you can push in a burst, and it refills continuously rather than on a date.
- The monthly pool is the total for the billing period. It resets when the period does.
A long autonomous run meets the window cap first; a heavy month meets the pool. Because they are separate, hitting one tells you something specific — a window stall means slow down, a pool exhaustion means the month was bigger than the plan.
Reaching a cap
Reaching a cap pauses further requests; it does not cancel your subscription or delete anything. A window cap clears on its own as the rolling window advances. A pool exhaustion lasts until the period resets, or until you add prepaid credit, or until you move to a larger plan.
Metered usage
Metered (pay-as-you-go) usage bills per request against a prepaid balance, with no window cap and no pool. It suits irregular or spiky work, and it is what API keys draw on by default.
A plan is not a discount on metered pricing — at current caps it is not cheaper per token. What a plan buys is a flat monthly spend and a larger burst ceiling. Pick a plan when predictability matters more than unit price; pick metered when your usage is genuinely occasional.