Limits

Calls are metered per app: every token an app holds counts against that app's limit in each store, 600 calls a minute unless the merchant set it otherwise for the app, with an optional quota per day or per month, counted in UTC. A publishable key's limit holds per visitor instead: each address calling with it (an IPv6 address by its /64) gets the key's calls a minute, so one visitor at it never holds back the others, though visitors behind one shared address share it too. A quota set on a key caps every visitor's calls together, and anyone holding the key, which is public, can spend it. A private file costs two calls each time: asking for its URL, and fetching it, and asking again past expires_at costs two more, though a video asked for with a publishable key holds four hours. Each range fetched is a fetch, a range a video element asks for on its own included, though fetches of one URL from one address within a few seconds of the first may share its call.

Past a limit, a call answers 429 with the reason limit, and Retry-After when the service knows when to retry.

Tokens a store refuses (unknown, expired, revoked or malformed) are also limited per network: past about 60 a minute from one address (an IPv6 address by its /64), or 600 across an IPv6 /48, calls to that store with a token the service hasn't accepted in the last few minutes answer 429 with the reason limit for up to a minute, never invalid_token, so keep the token and retry after Retry-After. A token in steady use keeps working. Publishable keys the store refuses count the same way, so a page whose key was revoked answers 429 for a while past that many calls; a key in use keeps working.

A call carrying an Origin header costs nothing more: reading which origins may read the answer doesn't count.