Odyssey has released Odyssey 3, an artificial intelligence model capable of building interactive virtual environments directly from text prompts. Unlike standard video generation tools, users can physically navigate and alter the simulated environments in real time. A research version of the model is currently open for public testing.
Odyssey built the system on an autoregressive diffusion transformer architecture. The software predicts physical interactions, object movement, and environmental changes over time. Odyssey 3 Pro scored 66.1 points on the Physics IQ Verified benchmark, representing the highest recorded public score. The model also secured top rank in 3 of the 4 test categories within the internal WorldMark benchmark.
Engineers tested Odyssey 3 as a training platform for physical hardware systems. The model generated training scenarios to control robotic arms, humanoid robots, and autonomous passenger cars. In 1 field test, the team trained an autonomous driving system for public roads in India using only 20 hours of driving data.
The software generates synchronized video feeds for multiple cameras at the same time. This capability allows autonomous AI agents to practice tasks within simulated environments based on natural language commands. Odyssey launched public access to a research version of the platform, allowing developers to test world generation and interact with the environments directly.
