Sume accepted job capacity by plan: 6, 24, 48, 120 and queue_full

How many jobs Sume accepts at once is the plan's concurrency plus its queue: 6 on Free, 24 on Pro, 48 on Startup and 120 on Scale and Enterprise. Another submit past that gets 429 queue_full, and it is safe to retry once a job completes.

This is a different limit from requests per minute, and it is the one that bites a batch.

Per plan

Numbers are from the generation admission page (read 2026-10-03).

Submit in waves

GET /v1/generation/admission-preview returns generation_limits, including wave_size_hint, which is the larger of 1 and the floor of 75 percent of the remaining queue capacity. It is a hint, not a promise. The in-flight budget is concurrency minus active minus queued, capped by the remaining queue capacity.

A 503 provider_capacity_exceeded is separate and is retried with the same idempotency key.

import math

def wave_size(queue_capacity_remaining):
    return max(1, math.floor(queue_capacity_remaining * 0.75))

for remaining in (5, 20, 100):
    print(remaining, wave_size(remaining))

Which error to expect

queue_full and rate_limited are both 429, so read the code, not the status. For a Free batch, six accepted jobs is the arithmetic, and for the provider side, see the 503 retry.