Sora Review

Judged as research, Sora is one of the moments this field will be dated by. The first clips shown in early 2024 held objects, faces and camera moves together across a shot in a way nothing else could, and the version that shipped the following year added synchronized sound and a likeness feature that dropped a real person into a scene on request.

Judged as a product, it is the clearest failure any large lab has had. Wrapping it in a scrolling feed turned a research triumph into a firehose of celebrity deepfakes, fabricated news footage and the specific flavor of junk that earned its own name, and the plan to make rights holders opt out of having their characters generated collapsed within days under obvious pressure.

The compute bill was enormous and the retention was not. OpenAI closed the consumer app in April and set September 24 as the end of developer access, folding the team into world simulation work aimed at robotics instead. What survives is the influence, which is everywhere, and a lesson about shipping a capability before deciding what it is for. The thinking and coding axes below describe a video generator that was never built to do either. There is nothing left here to buy.