250 prompts at 12s 1080p after Sora: which Sume models reach it
Four of the six candidate models make a 12-second 1080p clip on Sume, from $420.00 to $2,042.50 for 250 prompts. minimax-h3 and Omni Flash cannot.
For 250 prompts rendered as 12-second 1080p clips, wan-3.0 costs $750.00, kling-3 costs $420.00 with audio off or $630.00 with audio on, seedance-2-mini costs $1,277.50, and seedance-2-fast costs $2,042.50. minimax-h3 stops at 768p and gemini-omni-flash-1.1 stops at 10 seconds, so neither can produce this size in one clip.
This is the long, high-resolution corner of a re-render. OpenAI removed its Videos API and the sora-2 models on 2026-09-24 and listed no replacement, so a library that held 12-second shots has to find a model with that envelope.
Reach and price in one table
A clip counts as reachable only when the model accepts both the 12-second duration and the 1080p resolution. The price column is the billable cents for a single clip.
| Model | Reaches 12s 1080p | Per clip | 250 clips |
|---|---|---|---|
| wan-3.0 | Yes (2 to 30 s) | $3.00 | $750.00 |
| kling-3, audio off | Yes (4 to 15 s) | $1.68 | $420.00 |
| kling-3, audio on | Yes | $2.52 | $630.00 |
| seedance-2-mini | Yes (4 to 15 s) | $5.11 | $1,277.50 |
| seedance-2-fast | Yes (4 to 15 s) | $8.17 | $2,042.50 |
| minimax-h3 | No, 768p maximum | $0.90 at 768p | $225.00 at 768p |
| gemini-omni-flash-1.1 | No, 10 s maximum | n/a | n/a |
Why the Seedance prices are higher
Seedance is metered per 1,000 video tokens, not per second, and a 1080p clip carries many more tokens than a 720p one. At 12 seconds the 720p seedance-2-fast clip is 363 cents and the 1080p clip is 817 cents. Kling, in contrast, bills one rate for 720p and 1080p, so asking for 1080p there adds nothing: 168 cents either way when audio is off.
The Video Router docs also list minimax-h3-max, which reaches 1080p (a latent refinement from native 768p) for 5 to 15 seconds. A 12-second clip there is 240 cents billable, or $600.00 for 250. It is a different model from minimax-h3 with a different look, so test it separately.
A sensible split
If every shot really needs 12 seconds at 1080p, wan-3.0 and kling-3 are the low-cost pair in the catalog and seedance-2-fast is the premium line. If only some shots do, render those on a model from the table and render the rest at 8 seconds and 720p, where the library gets much cheaper.
The arithmetic is plain. Wan: $0.20 per second at 1080p times 12 is $2.40 list, times 1.25 is $3.00, which is 300 cents. Multiply by 250 and you get $750.00. Check the same numbers against GET /v1/videos/models before you submit a long batch; the Video generation docs describe the fields.
Sources
Related posts
More in Developers
- A 28-minute talk: three detach ranges, three STT requests, $0.31
Audio detach outputs up to 900 s but Sume STT takes up to 600 s, so a 1,680-second video needs three ranges: 3 x $0.01 detach + 28 x $0.01 STT = $0.31.
- 31 lookbook edits, 3 reference images each: 93 inputs, flat rows
31 edits at 2K with three references each cost $4.65 on Nano Banana 2.1 and $5.81 on Pro: Sume's rows are per image, not per input.
- 402 on shot 4 of 6: a $3.00 wallet and a mixed project, step by step
Walk a $3.00 balance through stills, voice, music, six shots and a render. See which submit returns 402, what was already paid, and how to resume safely.
- 402 on the third submit: six 5-second Wan 3.0 jobs on a $1.50 balance
Each 5 s Wan 3.0 720p job reserves $0.625 at submit. On a $1.50 balance the third POST /v1/videos fails with 402 insufficient_credits. Arithmetic and a script.
Written by Sume