Sora

Sora is a text-to-video generative model developed by OpenAI, announced on 15 February 2024 in the technical report "Video generation models as world simulators". The model was publicly released on 9 December 2024 as the Sora Turbo variant, available to ChatGPT Plus and Pro subscribers via sora.com.

Architecture

Sora is a diffusion transformer (DiT). Videos and images are represented as collections of spacetime patches, analogously to tokens in large language models. The model is trained in latent space (latent diffusion) and generates video through iterative denoising. The architecture is scalable — more compute translates into higher quality, longer and more consistent shots.

Capabilities

Sora generates clips up to 60 seconds long at resolutions up to 1080p, in multiple aspect ratios (such as 1:1, 16:9, 9:16). It supports three basic scenarios: text-to-video (video from a description), image-to-video (animation of an input image) and video-to-video (extending, blending and remixing existing clips). The model exhibits an advanced understanding of camera motion, multiple characters, physics and visual language.

Availability

Sora is available as a hosted product at sora.com and inside the ChatGPT app for Plus and Pro plan subscribers (with daily generation quotas and length / resolution caps that depend on the plan). The model weights are not publicly released. Generations are tagged with C2PA metadata and watermarks to indicate AI provenance.

Successor

On 30 September 2025 OpenAI announced Sora 2 — a next-generation model with improved physics, controllability and synchronised audio generation. Sora 2 is a separate model; this entry covers the Sora line in its first-generation variant (Sora 1 / Sora Turbo).