Grok Imagine Lite's 10 requests per second vs Sume plan concurrency

xAI lists a 10 requests per second limit and Batch API for Grok Imagine 1.5 Lite. On Sume, a clip batch is bounded by plan concurrency and queue capacity.

5 min readSume
All posts

On Sume, the limit that stops a video batch is not a requests-per-second number but accepted-job capacity: a Pro workspace can hold 24 paid generation jobs at once (4 processing plus 20 queued), and the 25th submit returns 429 queue_full. xAI's page for grok-imagine-video-1.5-lite lists a rate limit of 10 requests per second and Batch API support (xAI docs, read 2026-10-10).

Those are different controls, so a loop that worked against xAI's limit needs a different throttle on Sume.

The numbers on Sume

Sume's generation admission page separates four controls: generation concurrency, queue capacity, submit rate limits and balance reservation. Concurrency is a dispatch limit, not a submit limit: while queue capacity remains, valid jobs are accepted as queued. The default queue capacity is max(3, concurrency_limit x 5).

Concurrency by plan, Sume generation admission docs (repo, 2026-10-10)
PlanProcessingQueueAccepted jobs
Free156
Pro42024
Startup84048
Scale20100120

What a 100-clip Grok batch does

Take 100 image-to-video clips on the Grok row. On Pro, submit the first 24 and the rest must wait: the 25th submit gets 429 queue_full until a job finishes. On Startup the accepted limit is 48; on Scale it is 120, which holds the whole batch. Sume's docs say the dashboard Concurrency tab and the generation_limits.concurrency_limit field are the source of truth, not the static table.

A throttle that works

Treat queue_full as a signal to wait, not an error to retry in a tight loop. The errors page says to back off, use retry-after when it is present, and not to retry unsafe submits without an Idempotency-Key.

  • Keep at most accepted job capacity jobs in flight; submit the next when one completes.
  • Send an Idempotency-Key per clip so a retried submit returns the original job.
  • Use callback_url (HTTPS) so you are not polling 100 jobs.
  • Check 402 before the batch: the balance must cover each reserve.

Money check

Each Grok clip reserves list times 1.25 per second at submit: a 6-second clip is 6 x $0.0125 = $0.075, so a 100-clip batch reserves $7.50 if all run at once. xAI's batch price may differ; Sume's reserve is shown in the poll usage.cost when the job ends.

Sources

Related posts

More in Developers

All Developers posts

Written by Sume