How many 5-second AI video clips does $1 buy on Sume?

A dollar buys up to five 5-second clips on Sume, depending on the model and resolution. The cheapest is Gemini Omni Flash 1.1 at 360p, at $0.19 a clip for 5 whole clips. The most expensive per-second option in this table, Wan 3.0 at 1080p, costs $1.25 for one clip, so $1 does not cover even one.

How the number is built

Sume bills the provider list rate times 1.25 for each output second, and the catalog formula rounds up to the cent. A 5-second Wan 3.0 clip at 480p lists at $0.05 a second, so $0.25 list, $0.3125 with the margin, and $0.32 after rounding. Per dollar you get 3 whole clips, with $0.04 left over.

Why whole clips and not a ratio

You cannot buy 3.2 clips. Each job reserves a rounded amount, so the leftover cents stay in the balance. Each job rounds up by at most one cent, which is a bigger share of a cheap clip: Omni 360p at 5 seconds is $0.1875 before rounding and bills $0.19.

Using the table

If you only need a rough cut or a draft, start at 360p or 480p and re-render the keepers. Omni 360p at 5 seconds costs $0.19 against $0.94 at 1080p. Seedance models are priced per video token and are not in this table; read their rows in GET /v1/video-router/models. Confirm the rate for your model in the live catalog before you budget, because list prices come from the provider and can change.

Scaling the table to a budget

The table scales cleanly, with one caution. A $10 wallet buys roughly ten times the clips in the table, but not exactly. For Omni 360p, $10 is 52 clips at $0.19 with $0.12 left. For Wan 3.0 at 720p, $10 is 15 clips at $0.63 with $0.55 left. Rounding is per job, so the leftover is what the cents add up to.

Remember that failed jobs release their reservation, so a failed clip does not use up a slot in this count. A retried clip is a new job and a new reservation, so budget for your real keeper rate, not for one clip per prompt.

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.