Sora (app)
Last reviewed
May 16, 2026
Sources
25 citations
Review status
Source-backed
Revision
v1 · 3,618 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 16, 2026
Sources
25 citations
Review status
Source-backed
Revision
v1 · 3,618 words
Add missing citations, update stale details, or suggest a clearer explanation.
Sora is a short-form video application released by OpenAI on September 30, 2025, alongside the Sora 2 text-to-video model that powers it. The app combines a generative video tool with a vertical, algorithmically ranked feed similar in shape to TikTok and Instagram Reels, with every clip produced by an artificial intelligence model rather than recorded on a camera. Within five days of its invite-only debut on iOS, the app passed one million downloads and reached the number-one spot on Apple's United States App Store, becoming, by raw install velocity, OpenAI's fastest-growing consumer product after ChatGPT.
The launch marked the company's first deliberate push into a standalone social-media product. Where ChatGPT presented OpenAI as a productivity utility, the Sora app positioned the same lab as a consumer entertainment publisher, complete with a feed, profiles, follower counts, and a remix culture built around a feature called Cameos that let users insert their verified likeness into generated scenes. The release also pulled OpenAI into a series of fights it had largely avoided as a research lab: copyright disputes with Disney and other studios, deepfake complaints from actors including Bryan Cranston, and a backlash from the estate of Martin Luther King Jr. over racist depictions of the civil rights leader.
By early 2026 the app's commercial trajectory had reversed. Daily active use fell below half of its peak, 30-day retention collapsed to roughly one percent, and its App Store rating dropped to 2.8 stars on iOS. OpenAI announced in March 2026 that the consumer app and the sora.com website would close on April 26, 2026, with the underlying Sora API following on September 24, 2026. Bill Peebles, who led the Sora team, departed the company in mid-April 2026 alongside chief product officer Kevin Weil. The Sora app therefore occupies an unusual place in OpenAI's history: a record-breaking launch, an aggressive media cycle, and a short commercial life of fewer than seven months.
OpenAI first revealed Sora in February 2024 as a research preview, demonstrating a text-to-video diffusion model that could produce minute-long clips at 1080p resolution. That first version was never released as a public product. It opened in December 2024 as an experimental web tool, sora.com, available to ChatGPT Plus and Pro subscribers, but the company described it as a creator tool rather than a consumer destination. Through the first half of 2025, competitors including Google's Veo, Runway Gen-3, and Kuaishou's Kling closed the quality gap, and short-form video platforms continued to grow at the expense of text-driven social products.
The Sora app and Sora 2 were announced together on September 30, 2025, in a coordinated launch that paired a new model with a dedicated distribution channel. OpenAI framed Sora 2 as a step change in physical realism, audio synchronization, and instruction following. The accompanying blog post argued that prior video models often "warped objects and morphed reality" to satisfy a text prompt, while the new model would more often refuse a physically impossible action, producing, for instance, a basketball that rebounds off the rim when a player misses a shot rather than teleporting into the hoop.
The Sora app fit into a wider pattern of consumer experimentation at OpenAI during 2025. The company released the ChatGPT Atlas browser in October 2025, expanded the ChatGPT mobile app with a shopping surface, and rolled out the Apps SDK to put third-party experiences inside ChatGPT. The shared thread was a move away from a single chat interface toward several differentiated consumer endpoints. Sora was the most ambitious of these because it asked OpenAI to operate a social network with a content moderation surface, a recommendation system, and a creator economy, none of which it had run before. Internally, the project was led by Bill Peebles, a researcher who co-authored the diffusion transformer architecture and joined OpenAI in 2023 to lead Sora.
OpenAI released the Sora app for iOS on September 30, 2025, in the United States and Canada. Access was invite-only, with each new user receiving a small allotment of codes to share. The app required iOS 18.0 or later and a free ChatGPT account, and listed the app's developer simply as OpenAI. ChatGPT Pro subscribers could also access a higher-quality variant of the same model, branded Sora 2 Pro, through the web at sora.com.
In the first 24 hours, the app drew about 56,000 installs on iOS and reached number three overall on Apple's U.S. App Store. Daily downloads peaked at 107,800 on October 1, and by October 3 the app had passed ChatGPT to take the number-one spot on the chart. Bill Peebles, then head of the Sora team, said in a public post that the app had crossed one million downloads in under five days while remaining invite-only, faster than ChatGPT had reached the same milestone in late 2022. According to figures from analytics firm Appfigures, Sora drew roughly 627,000 iOS installs in its first United States week, edging out the 606,000 that ChatGPT had logged in its own first U.S. week.
The launch generated a heavy press cycle. Outlets including TechCrunch, The Verge, 9to5Mac, NBC News, and CNBC covered the release across multiple stories, focusing first on the technology, then on copyright behavior, and then on celebrity deepfakes. The combination of an invite-only feed, a novel Cameos feature, and OpenAI's existing brand attention pushed the app into the top of conversation for the first three weeks of October 2025.
The Sora app combined a generation tool, an identity system, and a social feed. The table below summarizes the main consumer-facing surfaces at launch.
| Feature | Description | Notes |
|---|---|---|
| Generation | Text-to-video creation powered by Sora 2, producing clips with synchronized audio, dialogue, and background sound | Clips were limited to about 10 seconds in the standard app, with longer durations in Sora 2 Pro on the web |
| Feed | Vertical, algorithmically ranked stream of user-generated and OpenAI-recommended clips | Layout closely resembled TikTok and Instagram Reels |
| Cameos | One-time video and audio verification let a user opt their likeness into generated scenes and grant friends permission to use it | Users could revoke permission and pull any video that featured them |
| Remix | Tap-to-remix tools that re-rendered an existing clip with a new prompt or a different Cameo | Helped drive viral chains, including the early Sam Altman deepfakes |
| Profiles and follows | Standard social graph features with public profiles, followers, likes, and comments | Followed the same general structure as short-video competitors |
| Sharing | Native share sheet plus a watermarked download for external posts | OpenAI confirmed users found ways to strip the watermark within days |
| Parental controls | Settings to limit feed time, restrict messaging, and control Cameo use for minors | Announced alongside the launch in response to expected scrutiny |
Cameos drew the most attention as a product distinction. The feature required a short verification recording, captured in the app, in which the user spoke a prompted phrase and turned their head to confirm liveness. Once verified, the user could star in any scene generated by themselves or by friends they had authorized. Sam Altman made his own Cameo available to all users, which seeded the feed with thousands of clips featuring an AI version of OpenAI's chief executive.
The early adoption numbers were strong by any historical comparison for an AI-native app. The figures below combine OpenAI's announcements with third-party analytics from Appfigures and reporting from TechCrunch, 9to5Mac, and The Vice.
| Metric | Value | Source / period |
|---|---|---|
| Day one U.S. iOS downloads | About 56,000 | September 30, 2025 |
| Peak day U.S. iOS downloads | About 107,800 | October 1, 2025 |
| First-week U.S. iOS downloads | About 627,000 | First week, Appfigures |
| Time to one million downloads | Under five days | Bill Peebles, public announcement |
| Top U.S. App Store rank | Number one overall | Reached by October 3, 2025 |
| Android day-one installs | Roughly 470,000 | November 4, 2025, TechCrunch |
| Android availability at launch | U.S., Canada, Japan, Korea, Taiwan, Thailand, Vietnam | November 4, 2025 |
| 30-day retention on iOS | Approximately 1 percent | a16z partner Olivia Moore, reported by Futurism |
| iOS App Store rating | 2.8 stars | December 2025, Vice |
The Android release on November 4, 2025 outpaced the iOS opening day in raw installs, partly because the iOS launch had built a backlog of demand and partly because Android shipped without the early invite gating in several countries.
The Sora app launched with an opt-out model for intellectual property, meaning rights holders had to actively ask OpenAI to block their characters from being generated. Within hours of launch, the feed filled with clips featuring Pikachu, SpongeBob SquarePants, Mario, characters from South Park, and likenesses of well-known fictional figures from Disney and Universal franchises. One widely shared clip depicted Sam Altman barbecuing a Pikachu, another showed SpongeBob cooking blue meth in a parody of the Breaking Bad television series, and a third placed Pokémon characters in pleading scenes asking Altman to stop training models on them.
On October 3, 2025, four days after launch, Altman published a blog post saying OpenAI would move Sora toward a more "granular," opt-in model in which rights holders could decide on a per-character basis how their intellectual property could appear, with revenue sharing on generated content under discussion. The about-face followed reporting that Disney and Universal had already sued the AI image platform Midjourney earlier in 2025 over similar issues, and that Disney had sent a cease-and-desist letter to Character.AI in the same week the Sora app launched.
In December 2025, OpenAI announced a three-year licensing partnership with Disney that, according to coverage in Futurism and Variety, would grant Sora and ChatGPT users the ability to generate clips using more than 200 Disney, Marvel, Pixar, and Star Wars characters starting in 2026. The deal reportedly carried a roughly one-billion-dollar value over its life. When OpenAI announced the Sora shutdown in March 2026, the Disney partnership was effectively abandoned. Coverage from CBC and Futurism reported that Disney executives learned of the decision less than an hour before the public announcement.
The table below summarizes the main copyright and likeness incidents during the app's lifetime.
| Date | Incident | Outcome |
|---|---|---|
| Sept. 30 to Oct. 2, 2025 | Feed fills with Pikachu, SpongeBob, Mario, South Park, and Disney character clips under opt-out policy | Studio backlash within days |
| Oct. 3, 2025 | Sam Altman announces shift toward granular, opt-in controls and possible revenue sharing | Sora restrictions tighten on commercial IP |
| Oct. 14, 2025 (approx.) | Estate of Martin Luther King Jr. requests block on MLK depictions after viral racist clips | OpenAI blocks MLK likeness in Sora |
| Oct. 20, 2025 | Bryan Cranston, SAG-AFTRA, UTA, CAA, and Association of Talent Agents press OpenAI on unauthorized voice and likeness use | OpenAI strengthens guardrails and publicly backs the NO FAKES Act |
| Dec. 2025 | Disney announces three-year licensing deal covering 200+ characters | Deal collapses when Sora is discontinued in March 2026 |
The Sora app rolled out in stages across platforms. Each rollout traded off the company's reluctance to widen the safety surface against the need to compete with established short-video platforms.
| Surface | Launch date | Initial regions | Notes |
|---|---|---|---|
| sora.com (web) | December 2024 (as Sora 1), refreshed with Sora 2 on Sept. 30, 2025 | Worldwide where ChatGPT was available, subject to subscription tier | Sora 2 Pro model available exclusively here for ChatGPT Pro subscribers at launch |
| iOS app | September 30, 2025 | United States and Canada | Invite-only, iOS 18.0 or later |
| Android app | November 4, 2025 | United States, Canada, Japan, South Korea, Taiwan, Thailand, Vietnam | Drew an estimated 470,000 day-one installs |
| Wider international rollout | Late 2025 through early 2026 | Selected European and Asian markets | Hampered by region-specific privacy and content rules |
| Discontinuation (web and apps) | April 26, 2026 | All regions | Sora API discontinued September 24, 2026 |
The web version retained more controls than the mobile apps throughout the product's life. ChatGPT Pro subscribers could use the Sora 2 Pro model on sora.com for higher-resolution and longer-duration clips, while mobile users were limited to the standard Sora 2 model and 10-second outputs.
The Sora app was conceived as a direct competitor to vertical short-video platforms, and reviewers and analysts compared it to TikTok, Instagram Reels, YouTube Shorts, and Meta's separate AI video app Vibes, released the same week. The table below summarizes the main structural differences.
| App | Owner | Primary content origin | Vertical short-video feed | Likeness / avatar system | Status as of mid-2026 |
|---|---|---|---|---|---|
| Sora | OpenAI | Generative video from Sora 2 | Yes | Cameos with one-time video verification | Consumer app discontinued April 26, 2026 |
| TikTok | ByteDance | User-recorded video and limited AI | Yes | Filters and effects, no built-in likeness platform | Active, global scale |
| Instagram Reels | Meta | User-recorded video with AI editing tools | Yes | Avatars and Meta AI characters | Active |
| YouTube Shorts | User-recorded video plus Veo-generated clips | Yes | Limited avatar options | Active | |
| Meta Vibes | Meta | Generative video from Meta's models | Yes | Cameos-like likeness inserts | Active |
Reviewers in the first month often noted that the Sora app's core technical demo was strong but the social product felt thin. Coverage from The Verge and TechCrunch argued that the feed could quickly grow repetitive because every clip was generated, the formats converged, and the lack of human performers and locations stripped out the variety that makes short video addictive. Olivia Moore of Andreessen Horowitz published widely shared retention figures showing that day-one retention was roughly 10 percent, day-seven retention near 2 percent, and 30-day retention close to 1 percent, an order of magnitude below mature short-video apps.
The Sora app raised three categories of regulatory and safety concern: deepfakes of real people, depictions of children, and copyrighted material. Each prompted concrete policy changes in the app's first month.
Within 24 hours of launch the feed contained widely shared deepfake clips of Sam Altman, including one in which an AI version of the OpenAI chief executive appeared to shoplift computer components from a Target store. The same week, advocacy organization Public Citizen sent OpenAI a letter calling for the app to be paused, arguing that Sora 2 made political and celebrity deepfakes effectively trivial to produce. CyberScoop reported on the letter on October 7, 2025. Safety researchers at the United States and United Kingdom AI Safety Institutes who had evaluated Sora 2 against OpenAI's red-team framework reportedly assigned the model a high-risk rating for misuse, although OpenAI proceeded with launch citing the watermarking and content controls in the app.
Following pressure from SAG-AFTRA, Bryan Cranston, United Talent Agency, Creative Artists Agency, and the Association of Talent Agents, OpenAI announced on October 20, 2025 that it had tightened restrictions on unauthorized voice and likeness generation and that it would publicly support the NO FAKES Act, a United States bill that would create a federal right of action against digital replicas of a person's voice or appearance. The estate of Martin Luther King Jr. separately asked OpenAI to block depictions of King after viral clips placed offensive and racist statements into his mouth; OpenAI applied the block in mid-October 2025.
The app's watermarking system also drew scrutiny. OpenAI added a visible animated watermark to downloaded clips, but reporters and security researchers documented within days that users were sharing tutorials on how to remove it, often with simple video-editing tools. The company acknowledged the issue publicly without proposing a technical fix beyond signaling that future revisions of the watermark would be more robust.
Reception split along an unusual fault line. As a software launch, the Sora app drew near-universal praise for its model quality, the polish of the Cameos verification flow, and the speed of the feed. As a consumer product, it drew increasingly negative reviews. By December 2025, BGR and Vice both ran stories noting that the app's iOS rating had fallen to 2.8 stars despite the record install numbers, with the most common complaints concerning repetitive recommendations, content quality after the novelty wore off, and the tightened restrictions on familiar characters following the October copyright reversal.
A TechCrunch report in late January 2026 found that downloads had fallen sharply and that OpenAI was scaling back marketing, although the company did not publicly acknowledge a strategy change at that point. By the time OpenAI announced the discontinuation in March 2026, reporting from Futurism and the European Business Review cited internal figures suggesting Sora cost roughly one million dollars per day to operate while serving fewer than 500,000 active users, down from a peak of about one million.
The shutdown drew a more divided reaction. Industry analysts cited by TechInformed and the Futurum Group argued that the closure signaled a broader retreat in consumer-facing creative AI, with OpenAI redirecting compute toward enterprise coding and agent products. Other commentators, including former OpenAI researcher Andrej Karpathy in a public post, argued that the Sora app had simply hit the structural problem that purely generative feeds get boring faster than feeds anchored in real people and events. Bill Peebles, in his announcement of his own departure on April 17, 2026, credited the app with prompting a wave of investment in video generation across the industry and called Sora "a project that could not have happened anywhere but OpenAI," while acknowledging in an earlier post that "the economics are currently completely unsustainable."
The Sora app is unlikely to be remembered primarily for its retention curve. Its more durable legacy is the Cameos pattern, the public debate over deepfakes that it forced into the open, and the way it pulled OpenAI directly into the social-media business for the first time, only for the company to step back out within a year.