Rate limits & concurrency

Stay within each key’s request rate and each account’s concurrency.

Limits

Limit	Applies to	Over it
60 requests a minute	Each key, across every authenticated `/v1` route. The public `/v1/models` catalogue isn't counted.	`429 rate_limit_exceeded`
5 generations at once	Your account, across keys and the Playground	`429 concurrency_limit`
Spending limit, optional	Each key	`402 key_ink_limit_reached`

Headers

Every 429 and 503 carries Retry-After: the seconds to wait before you retry.

`code`	`Retry-After`
`rate_limit_exceeded`	Until the minute frees a slot
`concurrency_limit`	15
`provider_unavailable`	5 or 60
`rate_limiter_unavailable`	5

Responses carry no rate counters. Track your own request rate.

Concurrency

A submit over the concurrent limit is refused: nothing is held and nothing runs. Queue on your side, and submit the next generation when a poll returns succeeded or failed.

Back off

Wait at least Retry-After, then retry.
If it repeats, double the wait each time, up to a minute.
Poll each generation every 5 seconds, no faster. Polls count toward the per-minute limit.

Explore models

loopling.ai · hello@loopling.ai · for agents · community · pricing · API · terms · privacy