Sora Is Gone: What It Means for AI Short Dramas

On March 24, 2026, OpenAI said in a post on X that it was “saying goodbye to Sora,” its AI video app, and that it would share timelines for the app and the API later. The news surprised the industry, NBC News reported, because the standalone Sora app, built on the second-generation Sora model with audio, had only launched the previous September and briefly topped Apple’s Photo and Video chart.

The shutdown then ran in two stages. According to OpenAI’s help page, as reported by The Decoder, the Sora web and app experiences closed on April 26, 2026. On the developer side, OpenAI’s own deprecations page says developers using the Videos API and the Sora 2 models were notified on March 24 of their removal from the API on September 24, 2026. The affected models include sora-2 and sora-2-pro and their dated snapshots, and the replacement column for every one of them is blank.

So as of mid-October 2026, both the consumer product and the API have passed their published shutdown dates, and OpenAI has not named a successor video model for developers.

OpenAI has said little in public. An OpenAI spokesperson told Engadget that the company was discontinuing Sora in the app and the API because of growing compute demand, and that the Sora research team would focus on world-simulation research for robotics. The rest of the explanation comes from reporting, not from OpenAI.

NBC News described Sora as resource-intensive and noted that moving compute away from it could free chips for coding, reasoning and text work, which NBC called more lucrative. It also cited The Wall Street Journal’s report that OpenAI executives said the company could not do “everything at once.” Engadget pointed to Appfigures data showing month-over-month declines in new installs and spending in early 2026. The partnership Disney announced in December 2025, which included a planned $1 billion investment in OpenAI, is not going ahead, according to NBC and The Hollywood Reporter as cited by Engadget. A Disney spokesperson said the company respects “OpenAI’s decision to exit the video generation business.”

Read together, the reporting points to a familiar trade-off: video generation is expensive to serve, and a company under pressure to focus spent its compute elsewhere. That is a reasonable reading of what was reported, not a statement OpenAI has made in full.

For anyone making or watching AI short dramas, the lesson is less about one company and more about where value sits. A vertical drama is a story told in one- to two-minute beats, each ending on a reveal. The model that renders a shot is one step in a longer chain of script, casting, storyboards, generation, editing, voice and subtitles. We cover that pipeline in how AI short dramas are made.

In that chain, the bottleneck has rarely been whether a model can make one striking clip. It is whether a production can keep the same heroine’s face and wardrobe across eighty episodes, keep the pacing tight, and do it at a cost per finished minute that a pay-per-episode business can carry. Tools vendors now pitch exactly those points. Google DeepMind’s page for Veo 3.1, for example, describes reference images of a character, scene or object to guide generation and says the aim is that characters keep their appearance across scenes. Whether any tool delivers that across a full series is something only a production can test.

TechCrunch’s panel discussion on March 29 framed the shutdown as a possible reality check for AI video: the claim that filmmaking by prompt was imminent ran into real technical and legal hurdles. The same piece noted that ByteDance had reportedly paused the global launch of its Seedance 2.0 generator earlier in March over engineering and legal questions.

Sora’s API gave developers six months’ notice and no named replacement. Any studio or app that had built its pipeline around one model’s look, prompt habits or API format had to re-tool on that schedule. That is the strongest argument for treating video models as interchangeable parts: keep scripts, character sheets and reference art in your own files, so a story can be re-rendered on a different model if the one you used disappears or changes its terms.

In practice, many teams already mix tools, using one model for a type of shot and another for a different one, and switching as quality and pricing move. We do not rank generators here; our guide to AI video generators for short dramas covers how to evaluate them. The point is that the asset worth protecting is the story bible and the characters, not access to any one model.

For audiences, the shutdown changes less than the headlines suggest. AI short dramas did not depend on Sora alone, and a viewer judges a series on whether the story holds up, not on which model drew it. If you are new to the format, what AI short dramas are explains how they differ from live-action micro-dramas. One practical effect is labeling: as productions move between tools, clear disclosure that a series is AI-made matters more than which tool made it, which our guide to AI-generated video labeling rules covers.

ShortFling does not generate video, and none of our vertical dramas can be watched yet. The drama catalog lists 12 series, all marked coming soon, and you can leave your email to be notified when one is released. We are not promising a date.

What is live today is fiction: serialized stories in SFW and NSFW sections, with the NSFW side behind an 18+ confirmation and availability that varies by region and platform. The Sora story is a reminder of why we start with story rather than any particular model: characters and plots outlast the tools used to render them.