4K AI video on Sume: Omni Flash $3.75 or an H3 upscale $2.00 for 10 s

Gemini Omni Flash 1.1 is the only Sume video model whose catalog lists 4K as a resolution: a 10-second 4K clip costs $3.75. MiniMax H3 is the other route to 4K pixels, but the Sume docs describe 2K and 4K there as upscales from native 768p, priced if your request asks for them, at $0.16 list per second for 4K. For 10 s that is $0.16 x 10 x 1.25 = $2.00.

Omni Flash price by resolution

Omni Flash prices per output second by resolution. The catalog list as of 2026-08-28 is $0.03 per second at 360p, $0.10 at 720p, $0.15 at 1080p and $0.30 at 4K, and it is subject to change.

Constraints on Omni Flash 4K

• Duration is 3 to 10 seconds, so 10 s is the longest 4K clip.
• Aspect ratio is 16:9 or 9:16 only.
• Native synced audio is always on, and the API rejects generate_audio: false.
• 4K is sent to the provider as 4k, and lowercase 4k is accepted as an alias on Sume.

The H3 upscale route

The Video generation page says that for minimax-h3 Sume bills 2K and 4K upscales if your request includes them. The model note lists $0.13 per second at 2K and $0.16 at 4K (list of 2026-08-25). The catalog's resolutions for H3 still shows only 480p and 768p, so read supported_resolutions from GET /v1/videos/models before you build on it. At the 1.25 factor, 2K for 10 s is $0.13 x 10 x 1.25 = $1.625, rounded up to $1.63, and 4K is $2.00. The catalog describes the upscale as starting from a 768p render.

Which one for 4K

If you need native 4K and can live with 10 s, 16:9 or 9:16, use Omni Flash at $3.75. If you need the cheapest large file and an upscale is acceptable, test the H3 route on one clip first. For comparison, Seedance 2.5 tops out at 1080p, where a 10 s 9:16 clip is $14.22. Pin the model id on Video generation rather than sending sume/auto, since Auto does not disclose which family runs and defaults to 720p.

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.