DBT Bureau
Pune, 2 Oct 2025
OpenAI has released Sora 2, its flagship video and audio generation model, marking a major step forward in AI-driven world simulation. The company describes this as the “GPT-3.5 moment for video,” surpassing the original Sora model launched in February 2024, which was widely seen as the “GPT-1 moment” for video generation.
Sora 2 introduces significant improvements in realism, physics, and controllability. The model can generate complex scenes such as Olympic gymnastics routines, backflips on paddleboards that account for buoyancy, and figure skating triple axels. Unlike earlier video models that often bent reality to fit prompts, Sora 2 is designed to obey the laws of physics—even modeling mistakes like missed basketball shots rebounding off the hoop.
The new system also handles multi-shot instructions, maintains world state across scenes, and supports cinematic, realistic, and anime styles. Alongside video, it generates audio, including speech, sound effects, and ambient soundscapes. A new feature called cameos allows users to insert themselves or friends into generated scenes, with accurate appearance and voice replication.
To showcase these capabilities, OpenAI is launching a social iOS app called Sora, powered by Sora 2. The app lets users create, remix, and share videos, with a feed designed to encourage creativity over consumption. Cameos are at the center of the experience, allowing users to securely place themselves into AI-generated content after a short setup.
OpenAI says it has built in safeguards around wellbeing, privacy, and teen safety, including parental controls, limits on teen usage, and stronger permissions for cameo sharing. The company is also deploying moderation teams and new recommendation algorithms that can be adjusted via natural language.
Availability: The Sora app is rolling out in the U.S. and Canada, with broader expansion expected soon. Sora 2 will initially be free with usage limits, while ChatGPT Pro subscribers will gain access to a higher-quality Sora 2 Pro model on sora.com. An API release is also planned, with Sora 1 Turbo remaining available.
With Sora 2, OpenAI is positioning video-audio generation as a cornerstone of future world simulation systems, potentially paving the way for AI models that better understand and interact with the physical world.
