Rate Limits
Requests per minute per key, and how to retry.
Requests per minute, per key
Each key can start 60 requests a minuteby default, counted in fixed one-minute windows. The limit applies to starting jobs and opening sessions. Uploads, job status and account reads aren’t counted.
Go over it and you get 429 with code rate_limited. There’s no Retry-After header yet, so back off and retry at the start of the next minute. Need more? The limit is set per key; ask us to raise it.
Platform capacity
Separately, there’s a ceiling on how much work the platform holds at once, and on how many live sessions run together. Hitting it returns 429 with at_capacity.
rate_limited isn't at_capacity
rate_limitedmeans you’re sending too fast: slow down. at_capacity means the platform is briefly full: retry shortly, at the same pace. They look alike and need different responses.