What Is Sora? OpenAI's Video Model — and Why the App Is Gone

Short version: Sora is OpenAI's video generation model. The standalone Sora app shut down on 26 April 2026, and the Sora API is scheduled to go dark on 24 September 2026. The Sora 2 model survives inside ChatGPT behind a paid plan. If you are reading a guide that tells you to download the Sora app, that guide is out of date.

I used Sora while it was live. Most articles ranking for this question still describe it in the present tense, which is how I ended up writing this one.

What is Sora?

Sora is a text-to-video model. You write a description, it generates footage. The second version, Sora 2, added synchronised audio, so clips arrived with sound rather than as silent video you scored yourself.

OpenAI shipped it two ways. There was the model, reachable through an API and inside ChatGPT. And there was the Sora app, a standalone social product with a feed of AI-generated clips that worked a lot like TikTok, except nothing in it was real.

The app's signature feature was cameos. You verified your own likeness once, and after that you could drop yourself into generated videos, and let other people do the same if you allowed it. Genuinely novel. Also the source of most of the trouble the app attracted, since a product built on inserting real faces into synthetic video is one bad actor away from a headline.

Is Sora still available?

The app: no. The model: yes, in one place.

What	Status	Date
Sora web + mobile app	Discontinued	26 April 2026
Sora API	Shutting down	24 September 2026
Sora 2 model in ChatGPT	Alive, paid only	Ongoing
Free-tier generation	Removed	January 2026

OpenAI announced the shutdown on 24 March 2026 and published a help-centre article, "What to know about the Sora discontinuation", which is the document to check if you need the official wording rather than mine.

That API date is the one that matters if you are technical. As I write this it is roughly five weeks out. Anything you built on the Sora API stops working in September, and "we'll migrate it later" is no longer a plan that fits.

Why OpenAI shut Sora down

OpenAI didn't say. No stated reason, which is itself a choice.

The numbers that are public tell a fairly blunt story. TechCrunch reported downloads peaking near 3.3 million in November 2025, falling to about 1.1 million by February 2026, and roughly $2.1 million in in-app purchases across the app's whole six-month existence.

Two million dollars. For a product generating video, which is one of the most expensive things you can do with a GPU. The revenue wasn't within an order of magnitude of the compute bill.

The more interesting failure is the social one. Sora bet that people would want a feed of AI-generated clips the way they want a feed of real ones. They didn't. Novelty pulled a few million downloads, and then the thing that keeps people returning to social feeds, that a real person they know did a real thing, was structurally absent. You cannot fake your way to that, because the faking is the product.

How Sora pricing works now

There is one route left: a paid ChatGPT plan. Plus sits at $20/month, Pro at $200/month, and Sora 2 Pro, the higher-quality variant, is the Pro-tier feature.

Pricing on the way down got messy, and if you find conflicting numbers online that is why. The app had a free tier, then didn't. Plus included Sora, then the app it was included in stopped existing. Several widely-shared pricing guides describe tiers that have not been purchasable for months. Trust OpenAI's own pricing page and nothing else here, including this paragraph, which will also age.

Is Sora free?

No.

It was, briefly. From 10 January 2026 OpenAI cut free accounts off from image and video generation entirely, and the free tier never came back. Now that the app is gone, the only door is a paid ChatGPT subscription. Anything promising free Sora access today is either out of date or reselling API capacity that expires in September.

What Sora was actually good at

Physical plausibility, mainly. Sora 2 handled the way objects move, collide, and settle better than most of what it launched against. Water behaved like water, cloth behaved like cloth. Video models fail in a distinctive way, where things drift and morph between frames, and Sora held together longer than its peers.

Native audio was the other real advance. Getting a clip back with matched sound removed a step that every other tool still leaves you holding.

What it was never good at was being a tool you could build a workflow on. Clips came out impressive and disconnected. Getting a consistent character across eight shots was a fight. And you cannot run a channel on a product whose access model changed three times in six months, which turned out to be a fair instinct.

What to use instead

Full disclosure before this section: Leaxor is mine. I've tried to be straight about where it doesn't fit.

Work out which half of Sora you actually wanted.

If you wanted cinematic clips, beautiful five-second shots to cut into an edit, then Kling and Google's Veo line are the nearest replacements. Both do the "generate a striking shot from a sentence" job Sora was known for. The Kling explainer covers what that costs.

If you wanted finished videos, clip generators are the wrong shape entirely, and this is the mistake I see most often. A raw clip is not a video. It has no script, no narration, no captions, and no continuity with the clip before it. Leaxor sits here instead: you give it a topic, and it writes the script, generates consistent characters, animates them, narrates, burns in captions, and returns a finished 9:16 MP4. Videos are 15, 30, or 90 credits depending on quality tier, credits are $0.10 each, and there is no subscription, so a standard video is $3.00. Different product, different job, and if you genuinely just wanted pretty B-roll then Kling is the better answer and I'd rather say so.

If you wanted the social feed and cameos, nothing replaces it. That was the part that didn't work commercially, so nobody is racing to rebuild it.

Should you still care about Sora?

As a product to adopt, no. It has a shutdown date on the only interface developers can reach, and the surviving route is a ChatGPT subscription where video is a feature rather than the point.

As a lesson, it's worth ten minutes. Sora was the most-hyped video model of its cycle, backed by the best-funded lab in the field, and it still could not turn a novelty into a habit. Model quality was never the constraint. Sora's clips were excellent right up until the day the app closed.

The thing to take from that: pick tools on whether they fit a workflow you will still be running next quarter, not on which demo reel looked best. Anyone who wired the Sora API into a production pipeline in January is spending part of this month ripping it out.

If you are making short-form video and want the whole job done rather than clips to assemble, that's what Leaxor's video mode is for. If you want a broader survey first, the alternatives roundup covers the field, and the Kling and Sora head-to-head has the model-quality comparison from when both were live.

Reading this with an AI assistant? Ask it to summarize this in ChatGPT, Claude, or Perplexity.