Sora is discontinued: what happened, and what to use instead

OpenAI announced the discontinuation on 24 March 2026. The consumer surfaces (sora.com and the Sora app) went dark on 26 April 2026, and the API follows on 24 September 2026. The two-stage shutdown gave developers a longer runway than consumers got.

The widely reported reason is economics rather than capability: Sora was expensive to run and never found revenue proportional to that cost, while usage fell away after the initial novelty. Reporting around the shutdown also points at unresolved copyright and likeness problems and a collapsed studio licensing deal. Whatever the precise weighting, the shutdown says something the benchmarks didn't: being the model that made the category famous is not the same as being a sustainable product.

Notably, Sora did not survive inside ChatGPT either. The integration that made it many people's first AI video tool went away with the standalone product, and OpenAI has announced no successor.

Sora's real strength was scene coherence and world-building: it held spatial consistency through camera moves better than most, so walk-through shots and invented environments survived a pan. If that's what you came for, Veo 3 is the closest current match on photorealism and it generates audio in the same pass, which Sora only gained late.

Its creation tools (remixing a generation with a text instruction, storyboarding multi-beat sequences) were ahead of the field. Nothing replaces them exactly. Runway is the closest in spirit: its toolset is built for directed iteration rather than one-shot generation, which is the same instinct.

Its distribution was the quiet advantage, and the thing that vanished most completely. "You already pay for ChatGPT, so you already have it" has no equivalent now. Veo 3 sits behind a paid Google AI plan, which is the nearest structural echo, but it's a separate subscription decision rather than a free bonus.

Its known weakness was physics and human mechanics, where hands and contact slipped uncanny in busy scenes. That was always Kling's territory, and still is.

For audio-synced, photoreal clips (the closest like-for-like): Veo 3. It generates dialogue, effects, and ambience with the picture, and leads on realism. It's the natural migration if Sora's audio was why you were there.

For realistic motion and physics on a budget: Kling. Recurring free daily credits mean you can evaluate it properly before paying, which Sora never offered.

For directed, iterative shot work: Runway, whose camera controls and Motion Brush give you the retake-and-adjust loop Sora's storyboard tool gestured at.

For no per-clip cost and no platform risk at all: the open-weight Wan line, which you run yourself. Worth noting after a shutdown like this one: an open-weight model you've downloaded cannot be discontinued out from under you. That's not a small consideration anymore.

It's worth sitting with the fact that the most famous model in AI video was switched off by the richest lab in the field. Not because it stopped working, but because the economics didn't hold. Every hosted generator you use runs on the same maths: GPU time is expensive, and the price you pay is set by a business that can change its mind.

The practical lesson isn't to avoid hosted models, they're better than what you can run at home. It's to know which of your work depends on a specific vendor staying interested. If a pipeline would break when a model disappears, that's an argument for open weights (the Wan line) or for keeping the prompt craft portable, since the habits transfer between models even when the models don't survive.