How to use the API Rate Limit
- Enter the API's limit and its window length.
- Enter your expected request volume over the same period.
- Add your concurrency or worker count.
- Read the required delay between requests and whether bursts will trip the limit.
How the calculation works
The sustainable rate is limit ÷ window seconds; the minimum spacing between requests is its reciprocal, multiplied up by the number of parallel workers so they share the budget. A 600-per-minute limit is 10 per second, so five workers each need a 500 ms gap. Ignoring concurrency is the usual reason a job that tested fine gets throttled in production.
Most providers implement a token bucket: tokens refill at a steady rate up to a bucket size, so short bursts above the average rate are tolerated until the bucket empties. Fixed-window limiters instead allow up to twice the limit across a window boundary, then block. When a 429 arrives, honour the Retry-After header; otherwise use exponential backoff with jitter, since synchronised retries from many clients recreate the spike that caused the throttle.
Rate = limit / window seconds ; delay per worker = workers / rateSource: Token bucket and fixed-window rate limiting algorithms; HTTP 429 and Retry-After per RFC 6585.
Worked example
Processing 90,000 records against a 1,200 requests-per-minute limit using 8 workers.
- Sustainable rate = 1,200 / 60 = 20 req/s.
- Total time = 90,000 / 20 = 4,500 s = 75 minutes.
- Per-worker delay = 8 / 20 = 400 ms between requests.
- Burst of 8 simultaneous calls consumes 8 tokens instantly, safe if the bucket holds ≥ 8.
About 75 minutes minimum, with each worker pausing 400 ms between calls to stay inside the limit.
Frequently asked questions
What does HTTP 429 mean?+
Too Many Requests. You exceeded the limit. Check the Retry-After header before retrying.
Why do I get throttled below the limit?+
Usually concurrency: parallel workers share one budget, so eight workers at the single-worker rate send eight times too fast.
What is jitter and why add it?+
Random variation in backoff delays. It stops many clients retrying in lockstep and recreating the spike.
Are limits per key or per IP?+
It varies. Check the provider's documentation, some apply both, and some count per endpoint.
Last reviewed August 31, 2026. We review this page whenever the underlying formula, tax year, published rate or standard changes.