We're pioneering world models, to go beyond language models
Odyssey is an AI lab focused on general-purpose world models: causal, multimodal systems that learn to predict and interact with the world over long horizons
We're pioneering world models, to go beyond language models
Odyssey is an AI lab focused on general-purpose world models: causal, multimodal systems that learn to predict and interact with the world over long horizons
Initializing Odyssey-2
Began simulating in
![](data:image/svg+xml,<svg display="block" role="presentation" viewBox="0 0 114 16" xmlns="http://www.w3.org/2000/svg"><path d="M 3 4.759 L 111 4.759 M 111 4.759 L 104 0 M 111 4.759 L 104 10 M 4 1 C 6.209 1 8 2.791 8 5 C 8 7.209 6.209 9 4 9 C 1.791 9 0 7.209 0 5 C 0 2.791 1.791 1 4 1 Z" fill="rgb(238, 238, 239)" height="10px" id="eYKSbJg_B" stroke-dasharray="" stroke-linecap="round" stroke-linejoin="round" stroke-width="2" stroke="rgb(238, 238, 239)" transform="translate(2 3)" width="111px"/></svg>)
A chicken running on a racetrack
Initializing Odyssey-2
Began simulating in
A chicken running on a racetrack
Initializing Odyssey-2
Began simulating in
A chicken running on a racetrack
World Model
Odyssey-2
Our most powerful general purpose world model yet, materially advancing the state-of-the-art in physical accuracy of world models
World Model
Starchild-1
A step beyond world models that learn only from visual observation, toward systems that learn from richer multimodal interaction with the world
World Model
Agora-1
A multi-agent world model, enabling multiple participants—human or AI—to share and interact within the same world simulation in real-time
Reinforcement Learning
PROWL
A novel RL-driven adversarial framework where an RL agent explores game environments with the objective to improve world model performance
World Model
Odyssey-2
Our most powerful general purpose world model yet, materially advancing the state-of-the-art in physical accuracy of world models
World Model
Starchild-1
A step beyond world models that learn only from visual observation, toward systems that learn from richer multimodal interaction with the world
World Model
Agora-1
A multi-agent world model, enabling multiple participants—human or AI—to share and interact within the same world simulation in real-time
Reinforcement Learning
PROWL
A novel RL-driven adversarial framework where an RL agent explores game environments with the objective to improve world model performance
World Model
Odyssey-2
Our most powerful general purpose world model yet, materially advancing the state-of-the-art in physical accuracy of world models
World Model
Starchild-1
A step beyond world models that learn only from visual observation, toward systems that learn from richer multimodal interaction with the world
World Model
Agora-1
A multi-agent world model, enabling multiple participants—human or AI—to share and interact within the same world simulation in real-time
Reinforcement Learning
PROWL
A novel RL-driven adversarial framework where an RL agent explores game environments with the objective to improve world model performance
What is Odyssey-2?
A world simulator that dreams in video
A world simulator that dreams in video
A world simulator that
dreams in video
Prompt with an image or text, and Odyssey-2 rolls out minutes-long interactive simulations
A Nascent World Simulator
Odyssey-2 is early, but it's learned from decades of video to simulate the world
Instant Streaming
Odyssey-2 begins an interactive stream instantly, generating frame-by-frame
Minutes of Streaming
Odyssey-2 streams multi-minute simulations, not 10 second clips
Applications of world models
A revolution for robotics
A revolution for robotics
For robotics, world models unlock intelligent robots capable of operating in general environments and serving billions of customers. World models learn how humans manipulate the world and enable robots to rehearse complex tasks before acting—simulating reaching, navigation, and manipulation learned from video. Instead of brittle automation in controlled settings, robots can adapt to all the variability of the real-world, like humans can
What makes Odyssey-2 different?
Odyssey-2 generates long-running simulations you can interact with
Traditional video models
Generating
ETA: Minutes
Generating
ETA: Minutes
A chicken running on a racetrack
A chicken running on a racetrack
Fixed 10 second clips
Takes minutes to generate
No interactivity
Odyssey-2
Generating
ETA: 50 ms
Generating
ETA: 50 ms
A chicken running on a racetrack
A chicken running on a racetrack
Generated in 50ms
Minutes of streaming simulation
Begins streaming instantly
Open-ended interaction
What will these models soon unlock?
General-purpose world models will soon proliferate interactive intelligence into the world
These models are nascent, but we see new software, hardware, and intelligence applications enabled in the near future
Intelligent guides for hospitality
Intelligent healthcare navigation and support
A reinvention of gaming
Accelerated robotic intelligence
Magic mirrors for instant try-on
Empathetic customer support intelligence
Personalized support for travelers
Educational health advertising
Accelerated robotic intelligence
Interactive advertising that engages
Interactive retail staff training
Live, immersive fitness software
Personalised operating systems
Intelligent guides for hospitality
Intelligent healthcare navigation and support
A reinvention of gaming
Accelerated robotic intelligence
Magic mirrors for instant try-on
Empathetic customer support intelligence
Personalized support for travelers
Educational health advertising
Accelerated robotic intelligence
Interactive advertising that engages
Interactive retail staff training
Live, immersive fitness software
Personalised operating systems
Realistic, controllable edge-case scenario generation
Intelligent, always-on brand ambassadors
Personalized, effective tutors for any topic
New types of interactive live events
Art that intelligently interacts with you
Personalized interactive training to upskill employees
New types of interactive games
New operating systems for home appliances
New types of simulations
Meaningful interactive companionship
Personalized fitness training
Realistic simulated training scenarios for warfighters
Realistic, controllable edge-case scenario generation
Intelligent, always-on brand ambassadors
Personalized, effective tutors for any topic
New types of interactive live events
Art that intelligently interacts with you
Personalized interactive training to upskill employees
New types of interactive games
New operating systems for home appliances
New types of simulations
Meaningful interactive companionship
Personalized fitness training
Realistic simulated training scenarios for warfighters
Research panel
The making of Starchild-1
Odyssey researchers discuss the core ideas and technical challenges behind Starchild-1
Research panel
The next frontier in AI
Hear from Odyssey researchers about why general-purpose world models are a world-changing technology
Research panel
The making of Starchild-1
Odyssey researchers discuss the core ideas and technical challenges behind Starchild-1
Research panel
The next frontier in AI
Hear from Odyssey researchers about why general-purpose world models are a world-changing technology
Research panel
The making of Starchild-1
Odyssey researchers discuss the core ideas and technical challenges behind Starchild-1
Research panel
The next frontier in AI
Hear from Odyssey researchers about why general-purpose world models are a world-changing technology
Our most recent news
What we've been working on recently
Our most recent news
What we've been working on recently
World Model
Odyssey-2
Our most powerful general purpose world model yet, materially advancing the state-of-the-art in physical accuracy of world models
World Model
Starchild-1
A step beyond world models that learn only from visual observation, toward systems that learn from richer multimodal interaction with the world
World Model
Agora-1
A multi-agent world model, enabling multiple participants—human or AI—to share and interact within the same world simulation in real-time
Reinforcement Learning
PROWL
A novel RL-driven adversarial framework where an RL agent explores game environments with the objective to improve world model performance
World Model
Odyssey-2
Our most powerful general purpose world model yet, materially advancing the state-of-the-art in physical accuracy of world models
World Model
Starchild-1
A step beyond world models that learn only from visual observation, toward systems that learn from richer multimodal interaction with the world
World Model
Agora-1
A multi-agent world model, enabling multiple participants—human or AI—to share and interact within the same world simulation in real-time
Reinforcement Learning
PROWL
A novel RL-driven adversarial framework where an RL agent explores game environments with the objective to improve world model performance
World Model
Odyssey-2
Our most powerful general purpose world model yet, materially advancing the state-of-the-art in physical accuracy of world models
World Model
Starchild-1
A step beyond world models that learn only from visual observation, toward systems that learn from richer multimodal interaction with the world
World Model
Agora-1
A multi-agent world model, enabling multiple participants—human or AI—to share and interact within the same world simulation in real-time
Reinforcement Learning
PROWL
A novel RL-driven adversarial framework where an RL agent explores game environments with the objective to improve world model performance
World Model
Odyssey-2
Our most powerful general purpose world model yet, materially advancing the state-of-the-art in physical accuracy of world models
World Model
Starchild-1
A step beyond world models that learn only from visual observation, toward systems that learn from richer multimodal interaction with the world
World Model
Agora-1
A multi-agent world model, enabling multiple participants—human or AI—to share and interact within the same world simulation in real-time
Reinforcement Learning
PROWL
A novel RL-driven adversarial framework where an RL agent explores game environments with the objective to improve world model performance