Luma Ray3.2 API prices are approximate; Sume reserves at submit

Luma's Ray3.2 API page puts a caveat on its own price table: prices are approximate and based on billing tokens. A 5-second clip at 1080p is listed at $1.20 and a 10-second clip at $3.60, so doubling the length triples the price. On Sume, the docs describe a different habit: the estimate is reserved from your balance at submit as provider list times 1.25, and usage.cost reports the billable amount afterwards.

What does the Luma table list?

The Build plan is pay per video with no minimum commitment, rate limits and no latency SLA. The Ray3.2 table lists text-to-video and image-to-video at 5 and 10 seconds across 540p, 720p and 1080p, video-to-video at the same resolutions, and Reframe priced per second. The page also says prices shown are SDR output, HDR output is 2x SDR pricing, and HDR plus EXR output is 3x SDR pricing.

Why is the 10-second price not double?

At 1080p the table gives $1.20 for 5 seconds and $3.60 for 10, which is $0.24 and $0.36 per second (arithmetic). The same pattern holds at 540p ($0.03 and $0.045 per second) and 720p ($0.06 and $0.09). Because the page calls prices approximate and tied to billing tokens, treat the second column of a long clip as a figure to confirm in your own account before you scale a job, not as a linear rate.

What does Sume say about quoting?

For video, the docs say the workspace USD balance is reserved on submit at provider list times 1.25 for every model, and usage.cost is the Sume billable amount. A request that cannot be covered gets 402 insufficient_credits before provider work starts. Limits differ per model: for example seedance-2.5 accepts 4 to 30 seconds and minimax-h3 5 to 15 seconds, and the docs tell you to read capabilities from GET /v1/video-router/models rather than assuming one envelope.

This post does not claim that Sume lists Ray3.2; check the live catalog for the ids it offers. The point is the quoting habit: ask the catalog for the model's pricing, compute the clip, and compare the reserve with what you expected.

What should you do before a large run?

Run one clip of each length and resolution you plan to use, and compare the charge with the table. If you plan HDR or EXR output on Luma, multiply by the stated 2x or 3x. On Sume, read usage.cost on the first jobs and keep an Idempotency-Key on submits that may be retried, so a retry does not create a second charge. Keep a small spreadsheet with the listed price, the observed charge and the difference for each length and resolution; after a handful of jobs you will know whether the approximation matters for your volume. On both services, a rate limit applies, so check it too before planning a batch.