Sora shut down: what happened and what to use instead

The short answer

According to OpenAI’s help page, Sora was shut down in two stages: the app stopped working on April 26, 2026, and the API, the interface developers and other services used to create Sora videos, stopped working on September 24, 2026. Since then Sora is not available, neither in the app nor through the API.

The job Sora did, a short video from a text description, is done today by several models available here with pay per generation: Veo 3.1, Kling 3.0, Seedance 2.5, Wan 3.0 and Gemini Omni Flash 1.1. All of them take a text description, all of them also take a start image and all of them can create sound together with the video.

For example, an 8 second clip at 720p with sound in Veo 3.1 Lite costs ₪0.707.03 crVeo 3.1 Lite, 720p · 8s · with sound right now.

What happened to Sora

OpenAI published a help page that explains the Sora discontinuation. According to it, the service closed in steps, first for app users and then for developers:

If you still have a question about old Sora videos or about your Sora account, the official answer is on OpenAI’s help page, and the link to it is in the sources at the end of this guide.

In practice it is simple: the prompts you wrote for Sora are still useful, they just need another model to run them. The models below take the same kind of description.

What happens to your videos when an AI service closes

The Sora case shows that a file kept only inside a service depends on that service. When the service closes, the files left in it close with it. So it is worth knowing where your creations are kept before anything changes.

Here every result is saved to our storage and to your personal library the moment it is ready. The file does not stay with the model’s maker, so it does not depend on it: even if a model stops being available one day, the files you already made with it stay in your library, and you can download them whenever you like.

• The first 1 GB of storage is free. Each extra GB costs ₪10 a month, in a small daily charge taken from your balance
• A model that is retired does not delete the files you made with it. They stay in your library, and you decide what to delete
• Credits are not tied to one model. If a model is retired, the same balance works in any other model, and it does not expire while your account exists

Which models do the job today

If you used Sora to make short videos from a description or from an image, these are the models that do that job here. The details in the table come from the makers’ official pages and match what you can create on the site.

In every one of these models you can choose a landscape clip, at 16:9, or a vertical one, at 9:16, which suits Reels and Stories. If price comes first, Google describes Veo 3.1 Lite as its most cost-effective video model, and it makes clips of 4 to 8 seconds with sound.

And if you want the highest ranked model: in an independent voting ranking for text to video, updated on September 21, 2026, Gemini Omni Flash 1.1 was first of 48 models. Google also recommends it as the replacement for the preview versions of Veo 3.1 in its developer API, which according to Google are set to shut down on October 22, 2026.

What a similar clip costs in each model

The table is computed as the page loads, with the same calculation the studio shows before every generation. Each row is one generation at the settings written in it.

How to choose by what you did in Sora

The quickest way to choose is to start from what you used most:

• A short clip with dialogue and atmosphere. The Veo 3.1 family creates dialogue, effects and ambience together with the picture. Try the idea in Veo 3.1 Lite or Veo 3.1 Fast, and move to Veo 3.1 for the final version
• A long clip in one go. Seedance 2.5 and Wan 3.0 reach 30 seconds in one generation, and Kling 3.0 reaches 15 seconds. Every option is in the guide AI video longer than 8 seconds
• Animating a photo. Every model in the table takes a start image, and some take an end image too. The full explanation is in the guide how to turn an image into a video
• The same character in several clips. Reference images keep the look: up to 3 in Veo 3.1, up to 10 in Wan 3.0 and up to 30 in Seedance 2.5. The details are in the guide how to keep the same character
• Comparing two leading models. If you are torn between Veo and Kling, one guide compares them head to head: Veo 3.1 vs Kling 3.0

Every model works from the same balance, so you can run the same prompt in two or three models and keep the result you like. When you switch the model in the studio, the exact price updates at once, before you send anything.

And what about Hebrew

None of these video models lists Hebrew on an official speech language list. Google writes about Veo 3.1 and Gemini Omni that English is fully supported, while other languages have not been evaluated and may work with varying results. Kling names five official speech languages for Kling 3.0: Chinese, English, Japanese, Korean and Spanish, and dialogue in another language is translated into English. We found no official speech language list for Seedance 2.5 or Wan 3.0.

So when you need a Hebrew voiceover, the safe way is to create it separately in a voice model whose official language list includes Hebrew, such as Eleven v3 or MiniMax Speech 2.8 HD, and add it to the video in editing. 1,000 characters of voiceover in Eleven v3 cost ₪0.474.69 crEleven v3, 1,000 chars right now.

The prompt itself is also best written in English for models whose maker says English is the language it evaluated. The full explanation, with the structure of a good prompt and a glossary of camera moves, is in the AI video prompt guide.

How to switch without starting from scratch

1. Collect the prompts that worked. A good prompt is the main asset left after Sora, and it works in other models too
2. Put them in a clear structure. Camera, subject, action, place and mood, plus dialogue and sound when needed. This structure comes from Google’s prompting guide for Veo 3.1
3. Test cheaply. Run the same prompt in two or three models, at a short length and a low resolution
4. Continue with the model that worked. Make the final version at the resolution and length you need, once the prompt clearly works

There is no subscription here and no plan to pick. You buy credits in advance, pay only for the generations you send, and choose for every generation the model that fits it exactly.

Questions and answers