Two 3-second clips or one 6-second clip: the rounding gap on Sume

On Sume, two 3-second clips cost slightly more than one 6-second clip of the same model and resolution, because each job is rounded up to the cent on its own. For Gemini Omni Flash 1.1 at 360p, two 3-second jobs bill $0.24 and one 6-second job bills $0.23. The difference is one cent, and it only shows when the per-second rate leaves a fraction of a cent.

The arithmetic

The catalog formula is list times 1.25, rounded up to the cent. Omni 360p lists at $0.03 a second, so Sume's rate is $0.0375 a second.

When it matters and when it does not

For one clip, a cent does not matter. For a batch it adds up: 1,000 three-second Omni 360p jobs bill $120.00 where the raw rate would be $112.50. Omni caps at 10 seconds, so you cannot always merge shots, but you can choose lengths that round less.

What to do with it

Look at the rate. At 720p Omni's rate is $0.125 a second, so an even number of seconds bills exact cents and an odd number rounds up half a cent: 3 seconds is $0.375, billed as $0.38. Check the cents with a script before a large batch, and use usage.cost on the poll response as the record of what was actually charged.

Where the gap changes the plan

The gap only decides things at the margin. If a 6-second shot works as one clip, one job is cheaper by a cent, and it is also one request, one poll loop and one file. If the same 6 seconds are two different shots, two jobs is the right plan, and the extra cent is the price of two prompts.

The larger decision is retries. A retried 3-second clip costs less than a retried 6-second clip. If your keeper rate is low, shorter jobs can cost less overall even with rounding, because you throw away less.

Before you ship anything, read the live pages again: the catalog is public, the pricing page is public, and the docs describe the request fields. A blog post is a snapshot. The catalog, the plan grid and the error table are the things that change, so write your code to read them instead of copying numbers from a page, and re-check when a new model is added.

A good habit is a small log line per submit with the model, resolution, duration, estimated cost, job id and the Idempotency-Key you used. When a job misbehaves, those six fields answer most of the questions support will ask, and they let you compare your estimate with usage.cost and the usage ledger without re-running anything.

If you are new to the API, start with one clip, one model and the lowest resolution, read the full response once, and only then build a loop around it. Most surprises with video jobs come from fields that were defaulted, not from fields that were set.