Odyssey-3 is a new generative world model that you can try for free
decoder ·
California-based AI company Odyssey has launched a public research preview of its new generative world model, Odyssey-3. This model generates interactive environments from text prompts in real time, allowing users to explore virtual spaces from various perspectives and trigger events. Built on an autoregressive diffusion transformer, the system simulates physical processes and predicts environmental changes based on user actions. While the base model features 14 billion parameters, the more powerful Odyssey-3 Pro supports higher resolution video generation. The company claims top scores on physics benchmarks, although these results come with caveats regarding testing methodologies and selection bias. Beyond entertainment, Odyssey-3 is designed to control robotic arms, drones, and video game characters by pairing the world model with specialized controllers. In demonstrations, the technology enabled robots to recover from failed grasps and allowed AI agents to autonomously play games like GTA V. Founded in 2023, Odyssey recently raised $310 million from investors including Amazon and AMD Ventures to further develop this spatial understanding technology.
decoder