California-based AI startup Odyssey has launched a public research preview of Odyssey-3, a generative world model. The system creates interactive environments from text prompts, allowing users to navigate them in real time. Developers can apply for API access to the technology.
The model utilizes an autoregressive diffusion transformer architecture to generate video frames based on user input. The base version contains 14 billion parameters and outputs at 832 × 480 resolution, while the Pro variant supports 1280 × 720 pixels.
Odyssey-3 achieved 66.1 points on the Physics-IQ Verified benchmark, though this result involved a selection method from multiple outputs. Without this selection, the model averaged 63.37 points across four standard runs.
The company targets applications in robotics, drones, and gaming. Odyssey is collaborating with Flexion on humanoid robotics and has demonstrated the model playing games like GTA V and Red Dead Redemption 2 using specialized controllers.
Founded in 2023 by Oliver Cameron and Jeff Hawke, Odyssey raised $310 million in June 2026 from investors including Amazon and AMD Ventures. The company aims to use the model to train AI agents in simulated environments.
