Luma Provisioned Throughput: 8-unit minimum vs Sume concurrency

Luma Provisioned Throughput sells capacity as requests per minute: 1 unit is 1 RPM on uni-1 and 0.4 on uni-1-max, with a minimum of 8 units. Sume works differently: generation concurrency is plan-only, and prepaid top-ups do not raise it.

Luma facts are from its Pricing page; Sume facts are from Generation admission and Errors and credits, read 2026-10-01.

How is Luma capacity priced?

What does the 8-unit minimum come to?

By the page's numbers, 8 units at the 1-month rate is 8 × $3,800 = $30,400 per month, at the 1-year rate 8 × $2,100 = $16,800 per month. Eight units give 8 RPM on uni-1 or 3.2 RPM on uni-1-max. The plans include a no-train guarantee, and video bills against a separate class you arrange with sales.

How does concurrency work on Sume?

The docs say generation concurrency is plan-only and prepaid top-ups do not raise the processing limit; admin overrides can. Queue capacity defaults to max(3, concurrency_limit × 5). There is no per-RPM purchase to compare with.

Concurrency is a dispatch limit, not a submit limit: jobs over the limit are accepted as queued while queue capacity remains.

What happens in a burst on Sume?

Jobs queue first. When concurrency plus queue capacity is full, submit returns 429 queue_full; request volume over the submit limit returns 429 rate_limited, to be retried with backoff and an idempotency key. See the Luma concurrency comparison for the shared-capacity side of Luma.