4K AI video API: which Sume models accept 4K and what a clip costs

Two models accept 4K in Sume's Video Router: gemini-omni-flash-1.1 and minimax-h3. A 10-second Omni clip at 4K is $3.75, and a 5-second H3 clip with a 4K upscale is $1.00. Seedance 2.5, Wan 3.0 and minimax-h3-max do not take 4k.

What the schema allows

In apps/api/src/schemas.ts the Router validator rejects resolution: "4k" with the message "4k is not supported by Video Router v1" for every model except those two. For H3, the Sume video models docs say Sume bills 2K and 4K upscales when the request includes them; native H3 output is 480p or 768p.

Comparing the two

Per second, Omni 4K is 37.5 cents and H3 4K is 20 cents, but they are different products: Omni clips run 3 to 10 seconds with audio always on, while H3 runs 5 to 15 seconds. The per-clip price is what you pay; the second is what you plan around.

A 4K clip is not a reason to skip a draft. Omni's 360p tier is 30 percent of its 720p price, so a draft costs a few cents (see the Omni 4K post). H3's two variants are compared in the H3 vs H3 Max post.

Not a 4K route

Seedance 2.5 tops out at 1080p ($42.65 for 30 seconds) and Wan 3.0 at 1080p ($7.50 for 30 seconds). If you need 4K from those, an upscale step is a separate decision; check the Video Router docs rather than assuming it is bundled.

• 4K: Omni and H3 only.
• H3 4K is an upscale and is billed as one.
• All prices include the 1.25 margin.

Budgeting 4K work

For a 30-second 4K deliverable, three Omni jobs of 10 seconds cost 3 x $3.75 = $11.25, and a 15 + 15 split on H3 with 4K upscales is 2 x $3.00 = $6.00. The H3 route is cheaper per second but yields clips of different character, so the choice is about the look you want first and cost second.

Keep the cheap drafting habit: generate at the native tier, review, and request 4K only for the keeper. Because the schema rejects 4k on other models, a request with the wrong model fails validation rather than silently downgrading, so you will see the mistake before any charge.

How these numbers were produced

I read the rate tables in the clone of Sume's repository on 2026-10-06 and called estimateGeminiOmniFlashProviderCost and estimateMinimaxH3ProviderCost for gemini-omni-flash-1.1 and minimax-h3, covering 4K and upscaled 2K jobs. The output fields are billableAmountUsdMicros and billableAmountUsdCents; the tables show the cents.

These are estimates of the billed amount for the stated settings. Anything not in the table, such as a different frame shape, a reference input or a changed quality tier, can change the number, so run the estimate for your own request before a large batch. Where this page and the public API pricing list disagree, the pricing page and the live quote win.