Odyssey Releases Odyssey-3 Foundation World Model Series, Tops Physics-IQ Benchmark
Odyssey introduced the Odyssey-3 foundation world model series, available in standard and Pro variants. The Odyssey-3 Pro model achieved a record-breaking score of 66.1 on the Physics-IQ Verified video-to-video benchmark. Accurate physical prediction is a major hurdle for generative AI models, which often produce visually realistic videos that violate basic laws of physics. Developing world models with high physical accuracy is essential for training autonomous vehicles, robotics control policies, and realistic interactive simulations. Built on an autoregressive diffusion transformer architecture, Odyssey-3 allows real-time generation and manipulation of embodied environments based on camera perspective shifts and user actions. The model also claimed top spots in the WorldMark benchmark across first-person stylized, third-person realistic, and third-person stylized environment evaluations.
## BACKGROUND
A foundation world model is an AI system trained on vast visual data to predict physical, dynamic, and causal relationships in an environment over time. The Physics-IQ Verified benchmark, co-developed by Google DeepMind, evaluates how accurately generative models can predict physical experiments across domains such as fluid dynamics, optics, and solid mechanics.