AI video cost per second at 720p on Sume: the models table

One second of 720p 16:9 video costs $0.5778 on Seedance 2.5, $0.3780 on Seedance 2.0, $0.3024 on Seedance 2.0 Fast, $0.1890 on Seedance 2.0 Mini, $0.1250 on Wan 3.0 and $0.1250 on Gemini Omni Flash 1.1. Grok Imagine 1.5 is the cheapest at $0.0125. These are the Sume amounts: provider list price times the 1.25 house margin, as the Video Router docs state.

The rows use different native sizes. MiniMax H3 and H3 Max render 768p, not 720p, and Kling 3 always renders 1080p, so they are in the lower table at their nearest tier.

Per-second price at 720p, list x 1.25

Seedance is billed per video token (width x height x seconds x 24 fps / 1024), so its per-second price depends on the frame size. The numbers below use the 1280x720 frame. Wan and Omni have flat per-second list prices, noted in the Video Router table with the fal read dates.

What a clip costs

The billed amount for a job is the list cost times 1.25, rounded up to the next cent. Multiply the per-second figure by your duration, then round up once per job, not per second. A 10-second 720p job is $5.78 on Seedance 2.5, $1.25 on Wan 3.0 and $1.25 on Omni Flash. The Omni Flash limit is 10 seconds, so that is its longest clip.

Duration windows differ: Seedance 2.5 takes 4 to 30 seconds, Wan 3.0 takes 2 to 30, Seedance 2.0 and its Fast and Mini siblings take 4 to 15.

Where the numbers come from

List prices and token rates are in the repo price tables (packages/provider-pricing) and the catalog docs, frozen here on 2026-10-13. The live source of truth is GET /v1/videos/models, whose pricing_skus already carry the x1.25. Re-read it before you commit a large budget; see the video generation docs for the schema.