Key points
- The model uses 14 billion parameters to create interactive worlds from text prompts.
- It achieves top scores on physics benchmarks, including fluid mechanics and optics.
- A free public research preview allows users to explore generated environments in real time.
California-based AI company Odyssey has launched a public research preview of its world model, Odyssey-3. According to The Decoder, this release allows users to try the 14-billion-parameter model for free. The system generates interactive environments from text prompts in real time. Users can explore these spaces and trigger events while the model responds instantly.
In plain English
Odyssey-3 creates virtual worlds that react to user actions. Instead of playing a pre-made game, users type a description to generate a new environment. The model simulates physical processes, predicting how objects move or change based on specific inputs. It learns cause-and-effect relationships from visual data. This allows the system to handle complex interactions, such as fluid dynamics or solid mechanics, within the generated space.
The background
Founders Oliver Cameron and Jeff Hawke first unveiled the model on September 15. That initial announcement focused on applications for robotics, autonomous driving, and video games. The Decoder reports that the new preview adds public access and more technical detail. The base model generates video at 832 × 480 pixels. A more powerful version, Odyssey-3 Pro, supports 1280 × 720 pixels.
What changes now
Developers can now apply for API access to integrate the technology. The free online demo runs on Odyssey-3 Flash, allowing users to choose between first-person and third-person perspectives. According to the company, Odyssey-3 Pro scored 66.1 points on the Physics-IQ Verified benchmark. This test evaluates physical behavior across multiple scientific areas. The methodology behind these results has some limits.
What to do and how to stay safe: Odyssey
- Review API access requests carefully before granting permissions to external services.
- Monitor compute resources for unusual spikes that may indicate unauthorized model usage.
- Ensure that generated content is not used in production environments without testing.
- Verify that data sent to the model does not contain sensitive internal information.
Step-by-step guide: How Software Updates Work: The Hidden Verification Steps
General security guidance from the Patch Herald newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Can Odyssey-3 control real-world robots?
The model is designed to serve as a foundation for controlling robotic arms and drones in simulated environments, not necessarily physical ones.
How does the model generate video?
It uses an autoregressive diffusion transformer that continuously generates new video frames based on previous frames and user actions.
What data was used for training?
Training data included internet videos with event descriptions, video game footage with inputs, and simulated physical interactions.
