OpenAI pulls the plug on Sora's consumer app this Sunday, April 26. The API follows in September. A model that launched barely over a year ago and had Disney committing a reported $1 billion in video work behind it is about to be a Wikipedia entry.
If you built a workflow on Sora, you already know. You've had a month to panic or port. If you didn't, the real story here isn't Sora. It's the shape of the AI video market now that one of the loudest names just conceded the field. What replaces Sora isn't one thing. It's five things, and none of them are interchangeable.
What actually happened
OpenAI announced the shutdown on March 24. The stated reason is cost. Video generation is the most expensive inference workload in the industry, and Sora reportedly ran well under the economics OpenAI needs ahead of a rumored IPO later this year. The consumer app goes dark April 26. The API sticks around until September 24, presumably to give enterprise customers time to migrate.
The secondary casualty is Disney's investment, which according to reporting was tied heavily to Sora's roadmap. That money doesn't vanish, but the product it was attached to does.
Where the market actually went
Here's the landscape Sora leaves behind, ranked by what each model is actually good for, not by benchmark scores.
Google Veo 3.1 is the technical high-water mark right now. It outputs native 4K at 60fps, generates synchronized audio in a single pass, and handles multi-shot coherence better than anything else you can type into. If you want a clip that sounds like it came from a camera, start here. Rumors of Veo 4 landing by late May are loud enough to plan around.
Kling 3.0 solved two problems Sora never did: length and price. Two-minute generations are routine where Sora capped out near 25 seconds. Cost per second is a fraction of what the big US labs charge. Kling is where a lot of creators quietly migrated three months ago and didn't bother telling anyone.
Runway Gen-4.5 is the most direct Sora replacement if you were using it for cinematic visuals. Its creative control surface — camera moves, shot continuation, reference-based generation — is the one Hollywood-adjacent editors actually reach for. Runway's CEO spent last week on a press tour arguing studios should fund 50 films instead of one $100M blockbuster, which tells you where they think the product is going.
Luma, Pika, and Hailuo each have their niche. Luma's motion is underrated and just got picked up for a new hybrid-filmmaking studio with Wonder Project. Pika's latest character consistency pass is better than most people give it credit for. Hailuo is where Chinese-language creators land by default, and the quality is not a compromise.
Six serious models. Each one better than Sora at something. None of them better at everything. That is the story.
The lesson is not "pick the next winner"
The thing to take away from the Sora shutdown isn't that you should swap one favorite model for another. It's that there is no next winner. Or rather, there's a different winner for each part of the pipeline, and the assignment shifts every few months as new versions release.
A sequence that uses Veo for the wide establishing shot, Kling for the dialogue-driven mid-section, and Runway for the cinematic close is a better sequence than one that uses any of them exclusively. That is not a theory. It is what serious AI filmmakers have been doing quietly since last summer.
The hard part is the orchestration. Keeping characters consistent across models. Keeping the score in sync with cuts made by three different video engines. Keeping the style coherent from shot one to shot forty. That is the problem worth solving, and it is the one Promvie is built around: the pipeline picks the right model for each stage and handles the seams so you don't have to.
If you are rebuilding a workflow this week, build it on that assumption. Any post-Sora stack that bets everything on Model X is a stack that is one IPO away from the same situation again.
What to do this week
If you have Sora clips sitting in a draft somewhere, export them before Sunday. If you are starting a new project, do not start by picking a model. Start by picking a pipeline and let the pipeline pick the models. If your tools do not let you do that, use tools that do.
A Sora-sized shutdown would have been a crisis a year ago. In 2026 it is an afternoon of config. That is the clearest signal yet that the center of gravity in AI video has moved from the model to the workflow.