Sora

What Sora was

Sora was a text-to-video model that OpenAI previewed on 15 February 2024, together with sample high-definition clips. The company said at the time that it could generate videos up to one minute long ^[1]. A text-to-video model takes a written description — a prompt — and produces a moving clip rather than a still image. The hard part is consistency: the people, objects and background have to stay recognisably the same from frame to frame, instead of drifting or melting as the seconds pass.

The name and the technology behind it

The name comes from the Japanese word for "sky" ^[1]. OpenAI described the technology as an adaptation of DALL·E 3 ^[1], the company's text-to-image system: the same broad idea of turning a text description into pixels, extended from a single frame to a sequence of frames.

In presenting the work, OpenAI said Sora could generate videos up to a minute long while maintaining visual quality and adherence to the prompt ^[3].

The larger pitch: simulating the physical world

OpenAI framed the project as teaching AI to understand and simulate the physical world in motion, aiming at models that help solve problems requiring real-world interaction ^[3]. That framing explains why a video model drew attention well beyond film-making. To keep a scene coherent over a minute of footage, a system has to pick up regularities in how things move, persist and interact — the kind of knowledge that simulation, robotics and other work on the physical world also depend on. Whether a model that produces plausible video has really learned such regularities, or only the appearance of them, is a separate question; the stated aim was the former.

Testing, a leak, and public release

At preview Sora was not generally available. In November 2024 a group of testers leaked API access on Hugging Face, and OpenAI revoked that access about three hours later ^[1].

Sora was released publicly on 9 December 2024 to ChatGPT Plus and ChatGPT Pro users in the US and Canada ^[1][2]. That is, it shipped first as a feature of a paid subscription rather than as a standalone tool. In February 2025 OpenAI announced plans to bring Sora into ChatGPT ^[1].

Place in the line

Sora was succeeded by Sora 2 on 30 September 2025, and OpenAI discontinued Sora in March 2026 ^[1]. The Sora page later carried a notice that as of 26 April 2026 the Sora product was no longer available ^[3].

The arc from preview to discontinuation took about two years. Within that span, text-to-video went from a research preview shown with sample clips to a feature inside a subscription product, and then to a next-generation model that replaced it.