Sora API shutdown alternatives: OpenAI lists no replacement
OpenAI's deprecations page removes the Videos API and sora-2 on 2026-09-24 and names no replacement. Pick a Sume video model by clip length and resolution.
There is no official drop-in replacement for the Sora API, because OpenAI's own deprecations page does not name one. The entry says developers were notified on March 24, 2026 that the Videos API and the Sora 2 aliases and snapshots are removed from the API on September 24, 2026, and its replacement column is empty. So the choice is yours, and the useful way to make it is by clip length and resolution, not by brand.
What the page lists as removed
Anything that still calls those ids has to move to another provider. Sume does not run an OpenAI video model, so the question is which model in the Sume catalog fits the clip you used to ask Sora for.
- The Videos API itself, so every endpoint under it.
- sora-2 and sora-2-pro.
- The dated snapshots sora-2-2025-10-06, sora-2-2025-12-08 and sora-2-pro-2025-10-06.
Sume video models by length and resolution
The table is read from the Sume video router catalog. Duration is the range one job accepts, in seconds.
| Model id | Duration (s) | Resolutions |
|---|---|---|
| seedance-2.5 | 4 to 30 | 480p, 720p, 1080p |
| seedance-2 | 4 to 15 | 480p, 720p, 1080p |
| kling-3 | 4 to 15 | 720p, 1080p |
| wan-3.0 | 2 to 30 | 480p, 720p, 1080p |
| minimax-h3 | 5 to 15 | 480p, 768p |
| minimax-h3-max | 5 to 15 | 480p, 768p, 1080p |
| gemini-omni-flash-1.1 | 3 to 10 | 360p, 720p, 1080p, 4K |
| grok-imagine-video-1.5 | 4 to 15 | 480p, 720p |
A short decision rule
Every job follows the same async envelope: submit with an Idempotency-Key, then poll the job status and fetch the result. Replace your Sora submit call with one of these, keep the polling, and re-check the clip duration your product assumed. Prompts will not reproduce Sora output, so re-approve the look before you switch traffic.
- Short social clip, 4 to 10 seconds: any row works, so choose on price and look.
- A single take longer than 15 seconds: only seedance-2.5 and wan-3.0 reach 30.
- Native audio in the same job: the catalog marks kling-3, wan-3.0, the H3 rows and the Omni row as audio-capable, so check
generate_audioon the row you pick. - 4K requested: only gemini-omni-flash-1.1 lists it as a resolution.
Migration steps in order
Do the mapping per call site because the models differ in more than price. A 6-second product loop and a 25-second scene do not belong on the same row.
- List every place that sends a sora-2 or sora-2-pro model id, including the dated snapshots.
- Record the clip length, aspect ratio and resolution each call site really uses.
- Map each call site to a row in the table above, not to one model for everything.
- Add an Idempotency-Key to every submit, since Sume requires it on writes.
- Render five prompts per model and have a person approve the look before cutover.
Sources
Related posts
More in Comparisons
- Seedance 2, Omni or Wan 3.0 for 9:16 ads on Sume: how to choose
Three Sume video models side by side for vertical ads after Sora: clip length, resolutions, 9:16 in the docs, references, and the price of a 5 s clip.
- Sync.so $0.05 a second vs Sume H3 Max lip sync: rates side by side
Sync.so lists $0.05 to $0.04 a second on four plans; Sume H3 Max lip sync derives $0.0625, $0.10 and $0.20. What each rate buys.
- Synthesia Starter's 12 minutes vs twelve 60-second Sume avatar jobs
Synthesia lists $29 for about 12 minutes and $89 for about 60. How those minutes map to Sume avatar jobs capped at 60 seconds, and what each setup is for.
- Tavus Starter vs Growth: break-even is about 1,014 minutes a month
Tavus Starter is $59 plus $0.37 a minute over 100; Growth is $397 with 1,250 minutes. The crossover, and when a rendered Sume clip fits better than minutes.
Written by Sume