IT之家 October 9 news: Odyssey posted a blog post yesterday (October 8) announcing the launch of the Odyssey-3 series basic world model. Its Odyssey-3 Pro achieved 66.1 points in the Physics-IQ Verified video-to-video benchmark test, setting the highest record on that list.
IT Home Note: The Foundation World Model is a general AI model that learns physics, dynamics, and cause-and-effect from extensive visual observations. It can predict how objects move, interact, and how environments evolve over time, and is used to simulate environments or train control strategies for systems such as robots and vehicles.
Physics-IQ Verified is a physical behavior benchmark test developed jointly by Anates Labs and DeepMind. It requires models to continue the real physics experiment videos and compare the predicted results with the actual results, covering fluid dynamics, optics, solid mechanics, magnetism, and thermodynamics.
In terms of models, the Odyssey-3 series basic world model is available in a standard version and a Pro version. The standard version balances physical accuracy with generation costs, while the Pro version offers stronger physical prediction capabilities.
In terms of architecture, Odyssey-3 adopts a auto-regressive diffusion transformer architecture, enabling the generation of embodied environments from prompts. After the user moves the view, takes actions, or introduces events during the generation process, the model combines existing observations and latest inputs to predict how the environment will change in real time.
This preview version supports first-person and third-person navigation, as well as independent camera movement. Users can move within the environment or trigger events to observe the model’s response to the environmental conditions.
In the WorldMark evaluation, Odyssey-3 ranked first in all three environments: first-person stylization, third-person realism, and third-person stylization, with scores of 77.2, 79.0, and 76.3 respectively.