California-based artificial intelligence company Odyssey launched a public research preview of its world model, Odyssey-3. Users can generate interactive environments from text prompts and explore them in real time. The model simulates physical processes and predicts how environments change based on specific actions. Developers can also apply for API access.
Founders Oliver Cameron and Jeff Hawke first unveiled the model on September 15 with a focus on robotics, autonomous driving, and video games. The latest release introduces public access, additional technical details, and benchmark results. A free online demo runs on the Odyssey-3 Flash version, letting users choose between first-person and third-person perspectives.
Odyssey-3 runs on an autoregressive diffusion transformer
Odyssey-3 uses an autoregressive diffusion transformer that continuously creates new video frames based on previous frames and user actions. Odyssey states that the model learns physical relationships and cause-and-effect from visual observations during training. Training data included internet videos with event descriptions, video game footage paired with keyboard and mouse inputs, and simulated physical interactions. An extra training technique reduces the required compute steps to enable real-time generation.
The base Odyssey-3 model contains 14 billion parameters and generates video at 832 by 480 pixels, according to the official benchmark submission. The Odyssey-3 Pro version supports 1280 by 720 pixels. Odyssey-3 Pro scores 66.1 points on the video-to-video benchmark from Physics-IQ Verified, according to the company. This test evaluates physical behavior across fluid mechanics, optics, solid mechanics, magnetism, and thermodynamics.
Benchmark claims come with caveats and high scores
The top score of 66.1 points involves a selection method that picked one of eight generated videos for each task. Benchmark rules require four test runs with standard deviation reported for any record claim, meaning the reported record does not meet that standard. Without the selection method, Odyssey-3 Pro averaged 63.37 points across four runs. Both results appear on the official leaderboard from Odyssey submissions.
On the WorldMark benchmark, Odyssey-3 ranks first in three of four categories based on internal evaluation: First-Person Stylized at 77.2, Third-Person Real at 79.0, and Third-Person Stylized at 76.3. In First-Person Real, it places third with 80.6 points. WorldMark tests how well models follow control instructions, image quality, and consistency over time.
Read nextOpen WebUI adds shared chat replies and model controlsThe model targets robotics and gaming applications
Odyssey aims to use the same world model across robotic arms, drones, and video game characters by pairing it with specialized controllers. In tests, an AI built on Odyssey-3 controlled multiple robotic arms after training on a few dozen hours of demonstration data. For humanoid robots, Odyssey is collaborating with Swiss robotics company Flexion. Odyssey also built a drone controller trained on simulated flight data.
The model can play video games, and the company demonstrated an AI autonomously playing GTA V. A controller trained on roughly two hours of GTA footage transferred its skills to Red Dead Redemption 2 without extra training. Odyssey plans to use the world model for training AI agents to learn from their own actions.
Odyssey was founded in 2023 by Oliver Cameron and Jeff Hawke. In June 2026, the company raised $310 million from investors including Amazon and AMD Ventures. Google DeepMind is pursuing a similar approach with Genie 3, and World Labs is developing comparable technology. The public research preview offers interactive environment generation for now, while robotics and autonomous systems require further work.
