What 25% adds to a fal bill: 100, 1,000 and 10,000 seconds of Wan 3.0

On Wan 3.0 at 720p, fal's list is $0.10 per second (read 2026-10-09) and Sume's price is $0.125. The extra 25% is $2.50 on 100 seconds, $25 on 1,000 seconds and $250 on 10,000 seconds. If you only need one model and have no use for the rest of what Sume wraps around it, going straight to the vendor is cheaper.

What are the exact numbers?

How many clips is that?

At 10 seconds each, 100 seconds is 10 clips, 1,000 seconds is 100 clips and 10,000 seconds is 1,000 clips. The difference per 10-second clip is $0.25.

What does Sume put around the model?

Per the docs: one API shape for the whole catalog (/v1/videos), one USD wallet, Idempotency-Key on retries, a standard signed webhook envelope, and the same job status URLs for every model. See Video generation. If you call Wan 3.0 from one script and nothing else, those are not worth 25% to you.

When does the gap stop mattering?

• When the clip is one step in a pipeline that also needs images, voice, captions or a timeline render, and you would otherwise run three vendor accounts.
• When you switch models by prompt result: the multiplier is the same on every row, so a model change does not add a new price rule.
• When volume is small: at 100 seconds the gap is the price of a coffee.

How do I decide if it is worth it?

Put a number on your own time. If wiring a second vendor account, key management and a second invoice costs you two hours, that is worth more than the $25 on a 1,000-second run. If you already run the vendor directly at 10,000 seconds a month, the $250 difference is a real line item and you should keep the direct account for that model.

Nothing here claims Sume is cheaper than a vendor's own list on a single model. It is not: the margin is the difference.