Docs / Rate Limits & Quota

How subscription quota is calculated and reset

Monthly subscriptions include a usage allowance. Both 5-hour and weekly limits apply. Usage is metered in weighted request units and depends on the token volume of each request.

How is the subscription allowance metered — what exactly is the unit?

In WEIGHTED REQUEST UNITS, not raw request counts: one request costs (input tokens + output tokens + 10% of cached-read tokens) divided by 29,000, rounded up, with a minimum of 1 and a cap of 50 units. A request that includes images costs at least 5 units, counting only the images newly uploaded in that request — images re-sent from earlier turns are not charged again. Cache reads count at 10%, writing to the cache is free, and a failed request costs 0. Units depend only on token volume, not on which model you call. Actual usage and reset times are shown in the Console; internal web searches add no extra allowance charge.

What you actually spend depends on the token volume of each request:

5-hour and 7-day windows

Paid subscriptions use separate fixed 5-hour and 7-day windows, timed from each window’s start. Usage for a window resets when it expires, and the next call opens a new window. Check the Console or usage API for your reset times. If either window has insufficient quota, further calls can receive a 429 rate-limit response. Failed requests consume no quota, and subscription overage charges are not added automatically.

A monthly subscription uses a subscription key, with supported models sharing the plan quota. Pay-as-you-go API billing uses a separate key and balance. Select the matching key type in the Console, then configure your client protocol. Topping up API credit does not increase subscription quota.

Compare workflows on the plans page. To see usage live in Claude Code: Usage in Claude Code.

What a 429 means

Need help? Email support@llmapi.pro · Home