Skylark Models

Limits & budgets

Read from the running service, not written down here.

Which limit actually binds

If you are signed in, it is almost always your account allowance. That is the only budget keyed on something you cannot choose, so a second machine, a fresh browser profile or a newly minted key all draw from the same pool. Every inference response restates what is left of it under credits.account, and a refusal arrives as account_budget_exhausted with a Retry-After.

Below that sit the limits that meter the open playground, where there is no account to key on. A key's own budget is the figure you were shown, and per-device budgets keep one busy client from starving others behind the same address. Of those, the one a caller cannot opt out of is the per-network budget, because it is keyed on the address block the request came from rather than on anything the client supplies.

Above all of those sits a global in-flight cap. The detection server is shared with production workloads, so this API takes only a fixed slice of it and returns busy rather than queueing without bound.

On this page