Sora Is Shut Down: What to Use Next for Vertical Video, Audio, API and Price

TWno1 Editorial · Published 2026-10-18

Credits and specs follow TWno1's live billing settings. Last checked: 2026-10-05

OpenAI shut down the Sora API on September 24, 2026, about five months after closing the Sora app.

So the question is no longer whether Sora works. It is what to do with the work you made there, and what to use instead. We are not going to name one replacement, and the reason is further down. Instead, here are the models still running, sorted by four needs, with dates on every number so you can match them to your project.

Timeline: 184 days from announcement to API shutdown

The app closed 33 days after the announcement. The API lasted 184. Press reports say OpenAI is moving compute toward enterprise coding tools and a combined app, and keeping Sora only as internal research.

The Veo row is there on purpose. Model retirement is not an OpenAI quirk. As of 2026-10-05, all three Veo 3.1 model IDs still carry a "-preview" suffix.

If you still have work on Sora

OpenAI's help page says you can export your content from sora.chatgpt.com/sunset, and that Sora data is permanently deleted after the export period. A search snippet of the same page says purchased Sora credits can be used for Codex. Our sources do not give the export deadline, so check OpenAI's page for the exact date and do not wait.

Quick reference: 10 video models still running

Prices below are official API list prices checked on 2026-10-05, at 720p or the closest tier so the rows compare.

Seedance 2.5 bills by video token. Using BytePlus's published formula, a 9:16 720p frame is 720 × 1280, so one second is 720 × 1280 × 24 ÷ 1024 = 21,600 tokens. At US$10.70 per million tokens, that is about US$0.231 a second.

Midjourney has video too, but its docs mention no public API, and it only animates a starting image: 5 seconds, extendable to 21. It does not fit this table.

Pick by need

Vertical drama with dialogue. The shortlist is Veo 3.1 Fast, Seedance 2.5 and Kling 3.0. Seedance 2.5 lists 11 prompt languages including Chinese and runs 30 seconds per clip. Kling 3.0 speaks Chinese, English, Japanese, Korean and Spanish. Veo's docs say English is fully supported and other languages "have not been evaluated". For non-English dialogue we would test the first two before Veo.

Cheapest drafts. Veo 3.1 Lite at 720p is US$0.05 a second with audio included. Wan 3.0 at 480P is also US$0.05, and Vidu Q3-turbo at 540P is US$0.035. Drafts are for blocking and motion. Resolution can wait.

Long single takes. Seedance 2.5 and Wan 3.0 both reach 30 seconds per clip. Kling 4.0 has been announced at 30 seconds, but as of October 5 it had no general release and no API price.

One account, many models. Runway's developer API resells Veo 3.1, Seedance, Wan 3.0, Grok Imagine and Gemini Omni Flash. Luma and Pika also sell other vendors' models for their own credits. Convenient, but channel prices vary a lot. On Luma's Plus plan, Veo 3.1 with audio works out to about US$0.84 a second, against US$0.40 from Google directly. That is 2.1 times as much.

The common advice misses the point: stop looking for "the" replacement

"Sora is gone, switch to X" just moves the same risk to a different vendor. We would do the opposite and keep your project assets independent of any one model.

Moving is costlier than it looks. Even on one vendor with the same seed, the docs promise only "similar" results, never identical. Switch vendors and your prompts need restructuring too; Google's Veo guide and BytePlus's Seedance guide recommend different prompt shapes. What travels intact is your script, your storyboard frames and your character reference images.

Our own production review points the same way. When we went through one 15-shot episode, the video model caused 38% of the 112 problems we logged. Re-running with a different video model did not help: in 13 of the 15 shots it made the same mistakes from the same storyboard frame. The ceiling is usually the storyboard, not the model's name.

The Monday migration list

1. Export your Sora work. Check OpenAI's help page for the deadline.
2. Write down what you used Sora for. Aspect ratio, audio or not, clip length, API or app. Cross off whatever fails in the table above.
3. Keep two vendors, not one. Run the same three shots on both and judge with your own footage.
4. Download as soon as a clip finishes. Veo keeps outputs for 2 days, BytePlus keeps task results for 7, and xAI returns temporary URLs. A vendor is not your archive.
5. Read the content rules first. Seedance 2.5 refuses reference images or videos containing real human faces. Veo limits person generation to the allow_adult setting in the EU, UK, Switzerland and MENA. Omni Flash editing of uploaded video is unavailable in the EEA, Switzerland and the UK.

Limits, and when TWno1 is not a good fit

• This table will age. Every cell reflects 2026-10-05. "Not found" means we could not find an official statement, not that the feature is missing.
• API list prices only. Consumer apps use subscriptions and credits with different math. They are out of scope here.
• No quality ranking. We have not run these 10 models on the same shots. The table answers "can it do this", not "which looks best".
• TWno1 will not replace a Sora API integration. Its public pages describe a short-drama workspace and do not mention a video API to call, so if you need an API, go straight to the vendors above.

Related reading: Veo 3.1 Lite vs Fast vs Standard: Which Tier for Drafts and Which for Finals; Kling 4.0 vs 3.0: 30-Second Clips, 10 Keyframes, and Whether to Wait; Seedance Draft Mode: When a 480p Preview Before 1080p Actually Saves Money.

FAQ

When did Sora shut down?

OpenAI closed the Sora web and mobile app on April 26, 2026, and shut down the Sora API on September 24, 2026. Press reports date the announcement to March 24, 2026, which makes 184 days from announcement to the end of the Sora API.

Can I still get my old Sora videos back?

According to OpenAI's help page checked on 2026-10-05, users can export Sora content from sora.chatgpt.com/sunset, and OpenAI permanently deletes Sora data after the export period. Check OpenAI's help page for the exact deadline and save your files locally as soon as possible.

What is the cheapest AI video API to replace Sora?

Based on official API list prices checked on 2026-10-05, Veo 3.1 Lite costs US$0.05 per second at 720p with audio included, Wan 3.0 at 480P (Singapore region) costs US$0.05 per second, and Vidu Q3-turbo at 540P costs US$0.035 per second. These suit drafts; for finals, check audio, aspect ratio and clip length against your needs.

Which models can replace Sora for vertical AI short drama?

As of 2026-10-05, models offering both 9:16 output and native audio include Veo 3.1 (8-second clips), Seedance 2.5 (30-second clips, 11 languages), Grok Imagine Video 1.5 (15-second clips) and Gemini Omni Flash. Kling 3.0 has native audio, but we could not find an official statement of its supported aspect ratios.

Should I pick just one Sora replacement?

No. Sora went from announcement to API shutdown in 184 days, and Google retired Veo 2 and Veo 3.0 on 2026-06-30. A safer setup keeps two working video vendors and stores your script, storyboard frames and character reference images yourself, so switching models means retuning prompts rather than starting over.

References

• OpenAI help: What to know about the Sora discontinuation (checked 2026-10-05)
• The Decoder: OpenAI sets two-stage Sora shutdown (checked 2026-10-05)
• Gemini API pricing (checked 2026-10-05)
• Gemini API changelog (checked 2026-10-05)
• Kling developer pricing (checked 2026-10-05)
• BytePlus ModelArk model pricing (checked 2026-10-05)
• Alibaba Cloud Model Studio: Wan 3.0 video (checked 2026-10-05)
• Vidu API pricing (checked 2026-10-05)
• xAI: Grok Imagine Video 1.5 (checked 2026-10-05)
• Luma API pricing (checked 2026-10-05)

• Model reviews