How many Format runs per minute can my Sume plan start?

Your Sume plan sets a write budget per minute: 120 on Free, 300 on Pro, 600 on Startup, and 1200 on Scale. Creating a Format run is a write, and reads get 40 times the write budget, so the create rate is the limit you will meet first.

Four response headers tell you where you stand: ratelimit-limit, ratelimit-remaining, ratelimit-reset, and retry-after on a 429.

These budgets are per key plan, not per Format, so every integration sharing a key shares the minute. Give each service its own key where you can, so one noisy job cannot starve another.

The numbers

The Formats docs give writes and reads per minute for each plan. Reads are 40 times writes, so the read column is 4800, 12000, 24000, and 48000. A bulk queue creates up to 100 runs from one request, but the items still start at the concurrency you set, from 1 to 16.

A quick check of the arithmetic: Pro at 300 writes a minute is 18,000 an hour, and Scale at 1200 is 72,000. A daily batch of a few thousand runs fits any plan if you spread it, and breaks Free if you do not.

Polling cost

Polling spends reads. A 30 minute run polled with a delay doubling up to 60 seconds takes about 35 status reads, so a Free plan's 4800 reads a minute supports a very large number of runs polled at once. In practice your creates and your cancels, which are writes, are the budget to watch.

If you can use webhooks, you will spend far fewer reads. Use a poll only as a safety net.

Handling a 429

When a call returns 429, wait for retry-after seconds before retrying, and spread the retry with jitter so many workers do not wake together. Use the same Idempotency-Key on the retry of a create so a request that did land is replayed, not duplicated. The Call a Format page lists the headers.

A bulk queue is the way to smooth a burst. Submit up to 100 items in one create, set concurrency to what your plan and providers can hold, and let the queue pace the runs. See Bulk runs.

Do not retry in a tight loop. A client that ignores retry-after and tries again immediately makes the next window worse, and a few of them together look like an attack.

Capacity planning

Estimate your peak create rate, not the average. A nightly job that creates 500 runs at once will exceed 120 writes a minute on Free in a few seconds, so batch it or pick a plan to fit. Track ratelimit-remaining in your logs, and alert when it falls below a tenth of the limit during normal traffic.

Review the budget each quarter. Usage grows quietly, and moving to a bigger plan is cheaper than the engineering time spent on tuning around a limit that no longer fits.