Sora AI Review (2026): What Happened to OpenAI's Video Generator

In October 2025, the No. 1 app on the US App Store was a video generator that let you drop your own face into a scene with a dragon, a talk show host or a cartoon dog. A year later, that app no longer exists. If you are searching for Sora AI today, here is the short answer: OpenAI closed the Sora app and website on April 26, 2026 and switched off the Sora API on September 24, 2026, so there is currently no way to make a new Sora video, free or paid.

That makes this an unusual review. We reviewed Sora as it stands in late October 2026, drawing on OpenAI's help documentation, independent hands-on reviews published while it was live, leaderboard data and user reactions. Nobody at The AI Future tested it for this piece. What follows is an account of what Sora was, how good it got, why it ended, and what you should use instead. There is also one twist: a report on October 12 says OpenAI is considering bringing the model back through a startup.

What is Sora AI?

Sora is the name OpenAI gave to its text-to-video models and, later, to a consumer app built around them. In plain English, you typed a description ("a corgi surfing at sunset, handheld camera") and the model produced a short video clip that matched it, eventually with sound.

The history matters, because "Sora" meant different things at different times. According to Wikipedia's timeline, OpenAI first previewed Sora in February 2024 and opened it to ChatGPT Plus and Pro subscribers in the US and Canada in December 2024. That first version, described by invideo's history of the product as Sora Turbo, made silent clips of up to 20 seconds at 1080p on sora.com.

The bigger moment came on September 30, 2025, when OpenAI launched Sora 2 alongside an invite-only iPhone app. Decrypt described it as a TikTok-style social app that "puts you in the videos". An Android app followed about two months later, and the sora-2 and sora-2-pro models reached the OpenAI API on October 6, 2025, per invideo.

So, who owns Sora AI? OpenAI built and owned it. The research was led by Bill Peebles, who, according to PYMNTS, citing The Information, is now reportedly a co-founder of a filmmaking startup alongside Jeffrey Katzenberg and Sujay Jaswa that is in talks to license the model.

Key features of the Sora AI video generator

These describe Sora 2 as it worked before the shutdown. None of them are available today.

Video with synchronised sound

Sora 2's headline improvement was audio. It produced dialogue, sound effects and background ambience in the same pass as the picture. At launch, Decrypt noted that Google's Veo 3 was the only other model doing this. OpenAI also claimed better physics: rather than bending reality to satisfy a prompt, Sora 2 would show a missed basketball shot bouncing off the backboard.

Cameos

Cameos were the feature that made the app spread. You recorded a short verification clip of your face and voice, and Sora could then place you in generated scenes. You decided who else could use your likeness and could revoke access later, according to Decrypt. In December 2025, Disney agreed to a three-year licensing deal that would let users generate more than 200 of its characters, alongside a $1 billion investment in OpenAI.

Clip length, storyboards and remixing

Clips in the app started at 10 seconds and later grew to 15, with Pro users on the web able to reach 25 seconds through storyboards, invideo reports. The API offered fixed lengths from 4 to 20 seconds and could chain extensions to about two minutes. The app also let you remix other people's videos, which fed its social feed.

The Sora AI watermark and provenance

Every video from the app and sora.com carried a visible, moving Sora AI watermark plus C2PA Content Credentials, a metadata standard that labels media as AI-generated. Videos made through the API carried the metadata but no visible mark. The watermark did not hold up for long: Wikipedia notes that 404 Media reported third-party watermark removers were widespread within a week of launch. A visible watermark works a bit like a "sample" stamp on a photocopy. It stops casual misuse, not determined misuse.

Is Sora AI free? Pricing and plans

Not any more, because there is nothing left to buy. The table below shows how Sora was priced in its final months, so that you can make sense of old guides and compare it with alternatives. We checked the current status and these historical prices in October 2026.

The free tier is the part people remember. In October 2025, OpenAI began selling extra generations at $4 for 10 and said the 30 free daily generations would shrink. Bill Peebles said at the time: "Eventually we will need to bring the free gens down to accommodate growth." API prices come from Omid Saffari's pricing breakdown, which also notes that an 8-second clip cost $0.80 on sora-2 and $5.60 on sora-2-pro at 1080p.

If you paid for Sora, OpenAI's help centre says purchased credits can still be used in Codex, and subscription refunds follow the normal ChatGPT refund policy. It also says Sora data will be permanently deleted once any final export window ends.

Performance and independent testing

While it was live, Sora 2 was a strong model, though not the best one. On the VideoGen-Eval arena, which ranks models by pairwise human votes, CodeSOTA's March 2026 snapshot placed Sora 2 Pro fourth with an Elo score of 1367 and standard Sora 2 ninth at 1342. Three versions of Google's Veo 3.1 sat above it, with the top one at 1381. In other words, people often preferred Sora's output, but they preferred Google's slightly more often.

Hands-on reviews from the time line up with that picture. Janine Heinrichs, writing for Unite.AI, found clips realistic and audio sync clean, with cameo setup taking about a minute. She also found longer videos lost consistency, wait times varied, some prompts were blocked without explanation, and her cameo looked like her in some clips but not others. Her conclusion was that Sora 2 "fits better as a concept tool than a polished editor."

The app experience was rougher than the model. In October 2025, Vice reported a 2.8-star App Store rating from about 4,400 ratings, driven by disappointing outputs and complaints about strict moderation of copyrighted characters. BGR pointed to invite-code frustration and a feed it called "really not a great experience."

Then the numbers turned. TechCrunch, using Appfigures data, reported that January 2026 installs fell 45% month over month to 1.2 million, with lifetime consumer spending of only $1.4 million across 9.6 million downloads. By late January the app had dropped to 101st in the US App Store's free chart.

Why OpenAI shut it down

Here it helps to separate what we know from what is reported. What we know: on March 24, 2026, OpenAI posted "We're saying goodbye to Sora" on X, and a spokesperson said the Sora team would focus on world simulation research for robotics, Engadget reported. Disney then walked away from its $1 billion investment.

What is reported but not confirmed by OpenAI: running costs of roughly $1 million a day and a user base that fell from about one million to under 500,000, figures invideo attributes to Wall Street Journal reporting. The Decoder linked the decision to OpenAI moving computing power towards coding tools and enterprise customers. Think of a restaurant that is famous for one spectacular, expensive dish few people order twice: eventually the kitchen space goes to the menu that pays the rent.

Could Sora come back?

Possibly, but not as the app you remember. The Information's October 12 report says OpenAI is in talks with a filmmaking startup that wants access to Sora to sell software and services. It is unclear whether the startup would receive the model's weights. This is a report from a single unnamed source, so we would treat it as a possibility rather than a plan.

Who it's for

Right now, Sora is for nobody who needs to make a video. There is no Sora AI app, no Sora AI video generator free tier and no API to sign up for. As of October 20, 2026, ChatGPT itself still has no built-in video generator either.

It still matters to two groups. Former users should check whether they exported their library through sora.chatgpt.com/sunset before deletion; according to invideo, exported videos remain yours to keep. Developers and businesses who built on the API need to have migrated already, and the episode is a useful lesson in not depending on a single provider's creative model.

Sora AI alternatives

Most of what Sora 2 did is available elsewhere, and in some cases the alternatives scored better when both were live.

• Google Veo 3.1 is the closest replacement. It led the same human-preference leaderboard where Sora 2 Pro ranked fourth, generates synced audio, and its Fast variant cost about $0.10 per second at 720p, according to Omid Saffari's comparison.
• Kling 3 suits multi-shot sequences and storyboarded scenes, which invideo recommends for the storytelling jobs people used Sora for.
• Higgsfield bundles several video models behind one creator-focused interface, which reduces the risk of one model disappearing. Read our Higgsfield AI review for the details, or our Leonardo AI review if you mainly need images with some motion.

Verdict

Sora 2 was, for a few months, one of the most capable and most fun AI video tools anyone had used: realistic clips, convincing sound and a cameo feature that made AI video personal. It was also expensive to run, weakly protected by a watermark that was easy to remove, tangled in copyright fights and never close to paying for itself.

Our rating reflects what you can do with Sora today, which is nothing. We have scored features and performance on how the model worked when it was available, but value and ease of use reflect a product you cannot open. If you want what Sora promised, use Veo 3.1 or a multi-model platform now, and keep an eye on whether the reported startup deal turns into something you can actually try. For writing better video prompts in any of these tools, our guide to better AI prompts applies just as well.

FAQ

What is Sora AI?

Sora AI is OpenAI's text-to-video model family and the social video app built on Sora 2. It turned written prompts into short, realistic clips with synchronised sound and let users insert themselves through cameos. OpenAI closed the app in April 2026 and the API in September 2026, so it is no longer available.

Is Sora AI free?

Sora used to have a free tier with daily generation limits, plus paid top-ups at $4 for 10 extra videos. Today it is neither free nor paid, because OpenAI has shut down the app, the website and the API. Any site offering a "free Sora AI video generator" now is using a different model under Sora's name.

How to use Sora AI?

You cannot use Sora AI any more. While it was live, you signed in to the Sora app or sora.com with an OpenAI account, typed a prompt or uploaded an image, and generated a clip, optionally with a cameo. Former users can only check for an export, and new users should try alternatives such as Google Veo 3.1.

Who owns Sora AI?

OpenAI created and owns Sora. It was led by researcher Bill Peebles, and OpenAI says the work continues internally as world simulation research. In October 2026, The Information reported that OpenAI was in talks to give a filmmaking startup, reportedly co-founded by Peebles and Jeffrey Katzenberg, access to the model.