Limits and Queueing

Nothing here is a per-key quota: a key is not rate limited, and a burst is not rejected for being a burst. The API is bounded by how many solves the platform runs at once, and everything above that waits in a queue. This page describes what that means in practice.

Results and Polling

• A finished result is kept for a short window (minutes), then it expires: a late /getTaskResult answers ERROR_TASK_NOT_FOUND. Read the result when it is ready; do not archive the taskId and expect to come back to it later.
• Poll /getTaskResult every few seconds, not in a tight loop. A solve takes from a couple of seconds to a minute depending on the product and on whether the site is busy.
• Every call answers with HTTP 200, success or failure: read errorId (and errorCode) rather than the HTTP status. A client library that only checks response.ok will treat every error as a success.

The Queue

Tasks are solved as slots free up. A task you just created is behind everything that arrived before it, so a slow answer usually means a busy platform, not a stuck task.

When the queue is long enough that a new task would wait far longer than the solve itself, /createTask refuses it immediately with ERROR_TASK_QUEUE_FULL and refunds the reservation. The task never reaches a worker. That is deliberate: a task that waits longer than a result lives could never be delivered.

In that situation, back off and come back: retrying in a loop only lengthens the queue. Errors and Retries has the per-code rules.

Concurrency

There is no per-key limit on concurrent tasks today. Many customers run large batches through a single key, and the queue absorbs them.

Two things bound a batch in practice:

• The platform's slots. Everything past the running solves waits.
• The private-beta products. The payload API caps concurrent sessions per key and answers concurrency_limit with a retryAfterMs; its own page documents it.

Requests

• proxy is required for every non-ProxyLess task type and ignored on a ProxyLess one. One clean token, no whitespace: see Proxy Format.
• websiteURL must be the page that renders the challenge, without credentials, on a public host. A private or reserved address in the URL or in the proxy host is refused with ERROR_BLOCKED_TARGET.
• Fields are validated before anything is reserved: a malformed request is answered at once and never charged.

Per Product