OpenAI releases Sora publicly

OpenAI made Sora, its text-to-video model, available as a standalone product to ChatGPT Plus and Pro subscribers, roughly ten months after it was first previewed to the public with sample clips but no access. The version released, internally called Sora Turbo, was reported to run substantially faster than the model shown in February. Plus subscribers, at $20 a month, and Pro subscribers, at $200 a month, received different generation limits and output quality, with Pro users given a higher cap and priority processing. Access launched only in the United States and Canada; OpenAI did not make Sora available in the UK, Switzerland or the European Economic Area at launch, citing regulatory requirements it had not yet met.

OpenAI built provenance signals into every output: each video carried a visible watermark, and all clips embedded C2PA metadata, an industry-standard cryptographic signature intended to let platforms and researchers verify that a video had been AI-generated. This followed the same pattern the company had used with system-card review ahead of release — OpenAI published a system card for Sora the following day, addressing risks including deepfakes, sexual content involving real or fictional minors, and disinformation.

The gap between the February preview and the December release illustrated a pattern that had become common with generative-media models: labs increasingly showed capability demonstrations well before public availability, using the interval for safety review, red-teaming and the negotiation of provenance standards, rather than shipping a model at the moment it was first shown to work. Sora arrived into a market that had moved in the interim — competitors including Runway, Kling and Luma had shipped commercial video-generation products of their own during 2024 — so the public release functioned as much as a catch-up to rivals as a debut of a wholly new capability.