15-second clip with sound: Kling 3 $3.15, Wan 3.0 $1.88, Mini $2.84

For a 15-second clip with sound, the cheapest row is MiniMax H3 at $1.13 (it renders natively at 768p, not 720p). Wan 3.0 at 720p is next at $1.88, then Seedance 2 Mini at $2.84 (9:16), and Kling 3 with its sound switched on is $3.15. Gemini Omni Flash 1.1 is out of this race, because it stops at 10 seconds.

The arithmetic

Sume bills the catalog list times 1.25, rounded up to the next cent. Kling 3 is the only one of these rows with a separate audio-on rate in the catalog: $0.112 per second silent and $0.168 per second with audio.

How audio is handled per model

• Kling 3: audio is a price switch. The sound premium is $3.15 - $2.10 = $1.05 for 15 s.
• Wan 3.0 and the Seedance 2.x models: the catalog lists audio as supported and shows one price row, with no separate audio-on rate.
• MiniMax H3 and Gemini Omni Flash 1.1: native audio is always produced and there is no generate_audio toggle, so a silent clip is not an option on those rows.

Limits that matter at 15 seconds

Kling 3 accepts 4 to 15 s, so 15 s is its ceiling, and it offers 16:9, 9:16 and 1:1 only. Wan 3.0 accepts 2 to 30 s, so a 15 s clip is half of its range. Seedance 2 Mini is capped at 15 s as well. MiniMax H3 accepts 5 to 15 s at 480p and 768p. Check the exact envelope for any model with GET /v1/videos/models before you submit; limits differ by model. If you plan a clip longer than 15 s with sound, Wan 3.0 is the only row in this table that accepts it in one job (up to 30 s), and its 30-second price follows the same per-second arithmetic: 0.10 x 30 x 1.25 = $3.75.

Which to choose

Choose MiniMax H3 when price is the only constraint and 768p is enough, and Wan 3.0 when you want a true 720p row and room to go past 15 s. Do not read a price gap as a quality gap: the catalog does not rank output quality. Choose Kling 3 when you specifically want that model with sound on, and accept the $1.05 premium over silent. Choose Mini if you want the Seedance request shape (21:9 through 9:16, reference audio) at a middle price. See Video generation for the request fields.

Check it before you run it

Every figure above is a catalog list price times 1.25, rounded up to the cent, as of 2026-10-08. Catalogs change, so before a large batch, read the current model entry in the docs and recompute the one line that matters for your case. Write the arithmetic next to the job in your own notes: list rate, seconds or characters, multiplier, rounding. If the result differs from the wallet charge by more than a cent, the catalog entry has changed, and the docs page is the place to find out why.

Run one small job first. Submit a single request with the pinned model id and the settings in the tables, poll the returned polling_url until it finishes, and compare the charge with your estimate. Then scale up. Using a pinned id for the test matters, because sume/auto never names the family, so you cannot tie its charge to the row you priced.