1080p per-second vs per-clip at 7, 13 and 21 seconds: who accepts each

Length decides which models you can even price. At 1080p, a 7-second clip fits Wan 3.0, MiniMax H3 Max, Gemini Omni Flash 1.1, Seedance 2 and Seedance 2.5; a 13-second clip drops Omni Flash (10 seconds maximum); a 21-second clip leaves only Wan 3.0 and Seedance 2.5.

Per-second rates mislead here for one reason: they hide the length limits. The table gives both the clip total and the per-second rate, and marks models that refuse the length.

Per clip and per second

Prices from the price tool at 1080p; Seedance rows are 9:16. Limits from the docs: Wan 3.0 2 to 30 s, H3 Max 5 to 15 s, Omni Flash 3 to 10 s, Seedance 2 and 2.5 start at 4 s with maxima of 15 s and 30 s.

Read the shape of the table

Wan 3.0 and H3 Max are flat per second, so doubling the length doubles the bill. Token-priced Seedance rows scale the same way here but start from a much higher rate: Seedance 2.5 at 1080p is $1.4216 per second against $0.25 for Wan 3.0.

H3 Max's 1080p is a latent refinement of a native 768p render, which the docs state; Wan 3.0 and Seedance render 1080p natively in the catalog. That is the sort of difference a price table cannot show.

Gotchas

• A length outside a model's range returns a validation error (400 invalid_request), not a clipped video.
• Use supported_durations from GET /v1/videos/models in code rather than a hard-coded table.
• Longer is not always cheaper per second: no model in this table lowers its rate with length.