Odyssey Has Released New World Model
The Odyssey-3 model enables robots to understand complex physical environments through video-to-video learning.
Updated on Oct. 9, 2026 in Robotics

Live Poll
Do you believe advanced world models will improve how AI operates in the real world?
Odyssey has launched a research preview of its Odyssey-3 world model, designed to control various systems including robot arms, cars, and drones. The model utilizes extensive visual data to provide machines with a foundational understanding of the physical world.
Why it matters
By learning to predict scene changes from visual data, Odyssey-3 allows machines to grasp general environmental concepts before mastering specific tasks. This capability marks a significant shift in how AI-driven robots and vehicles interpret their surroundings.
Odyssey-3 Pro achieved a score of 66.1 on the Physics-IQ Verified benchmark. The system demonstrated its driving capability using only 20 hours of training data on roads in India.
The players
Odyssey
Odyssey is an AI laboratory based in Palo Alto that develops advanced world models for robotics and autonomous systems.
Flexion
Flexion is a technology firm that utilizes Odyssey-3 to create control policies for humanoid robots.
AMD
AMD is a major semiconductor company that reached an agreement to acquire World Labs for $8.2 billion.
World Labs
World Labs is a research organization that entered an acquisition agreement with AMD in September 2026.
The details
The model, currently available as a research preview to developers, processes robot demonstrations to learn and execute complex behaviors. Companies like Flexion are already integrating the technology to develop control policies for humanoid robots.
Timeline
June 2026: Odyssey raised $310 million in funding.
September 2026: AMD agreed to acquire World Labs for $8.2 billion.
October 2, 2026: Odyssey announced the research preview of Odyssey-3.
October 8, 2026: The news was published.
The Tech Race
The Odyssey-3 model's performance on the Physics-IQ Verified benchmark sets a new standard for robotic vision, just as similar benchmarks established the performance ceiling for previous generations of language models. This release positions Odyssey at the forefront of the shift toward general-purpose artificial intelligence in physical robotics.
Developers can now access the Odyssey-3 research preview to begin experimenting with these advanced control capabilities. This technology may eventually lead to more capable consumer robots and smarter autonomous vehicles that require significantly less training time.
The takeaway
Odyssey-3 highlights the growing ability of AI to learn complex physical tasks from minimal training data. This advancement suggests that future robotic systems will likely become faster to deploy and more adaptable to real-world scenarios.
Further reading
Explore the latest developments in automated hardware in the Robotics section.
Live Poll
Do you believe advanced world models will improve how AI operates in the real world?










