Re-render 100 Sora prompts at 8s 720p 9:16: six Sume models priced

Re-rendering 100 stored prompts as 8-second, 720p, 9:16 clips costs between $60.00 (minimax-h3 at 768p) and $242.00 (seedance-2-fast) on Sume, with wan-3.0 and gemini-omni-flash-1.1 both at $100.00. Every figure below is the provider list price times 1.25, rounded up to the next cent per clip.

OpenAI's deprecations page lists the Videos API and the sora-2 family (sora-2, sora-2-pro and three dated snapshots) as removed on 2026-09-24, with developers notified on 2026-03-24 and no replacement listed. A prompt library that was written against sora-2 therefore needs a new target, and the first question is what the 100 clips cost on each candidate.

The price of one clip, then of 100

The catalog price of a clip is fixed by model, resolution and length. Aspect ratio does not change the price for the models below, except that Seedance squares (1:1) are slightly cheaper at 480p, which does not matter here. Kling is the one model that prices audio separately.

How the numbers are built

Take wan-3.0. Its list price is $0.10 per second at 720p. Eight seconds is $0.80 list, and times 1.25 that is $1.00, so 100 cents per clip and $100.00 for 100. Take seedance-2-fast: the catalog meters it per 1,000 video tokens, and the 8-second 720p 9:16 clip lands at 242 cents, so 100 clips are $242.00.

minimax-h3 has no 720p tier. Its native tiers are 480p and 768p, and 768p at $0.06 per second gives $0.48 list, $0.60 billable. If your downstream player needs exactly 720p, treat 768p as a different size and test it before you count the $40.00 you save over wan-3.0 on 100 clips.

What the cheapest line leaves out

Price is one column. The models differ in what they accept, and a library that uses reference images or end frames may rule some of them out. The Sume video docs list the capability set of each model, and GET /v1/videos/models returns it for your key, so read it before you commit a whole library.

Kling-3 audio is a real choice: at 8 seconds the audio-on clip is 168 cents against 112 cents with audio off, a difference of $56.00 per 100 clips. Gemini Omni Flash and MiniMax H3 always produce audio, so they have no cheaper silent option.

• Run 5 prompts first on each finalist, not all 100.
• Log the usage.cost of each completed job and compare it to this table.
• Pin a model id for a library you want to keep reproducible; sume/auto is for when you do not mind the family changing.

Where to go from here

Once you pick a model, the code change is small: a new base URL, a Sume key, and a bare model id. See the POST /v1/videos docs for the request fields and the Video Router page for per-model limits.