How much does Odyssey cost?

No pricing is published on the site. Access is available through free "try" demos and by requesting API access, with terms presumably determined during that process.

Is Odyssey open source?

The models are presented as proprietary products with demo and API access; no open-source licensing is indicated on the site.

Who is behind Odyssey?

The Odyssey team brings experience from DeepMind, Waymo, Tesla, Meta, Apple, and xAI, and the company has raised a $310M Series B.

What can Odyssey's world models be used for?

The company positions its models as a shared foundation for robots, humanoids, cars, drones, and game-playing agents, with target domains including robotics, self-driving, drones, defense, gaming, energy, cyber, and AI training.

Odyssey — AI world models that understand and simulate reality

Odyssey is an AI lab building general world models — causal, multimodal systems that learn to understand and simulate the world. Its foundation world models capture physics, dynamics, cause-and-effect, and agent behaviors, providing a shared foundation for robots, autonomous vehicles, drones, game agents, and AI training. The company offers Odyssey-3 as its flagship foundation world model, Agora-2 for multi-agent shared simulations, PROWL-2 for reinforcement learning, and Starchild-1 for learning through multimodal interaction.

Key Features

Foundation world model (Odyssey-3)

Draws on a learned understanding of physics, motion, and cause-and-effect to act in physical and virtual worlds; one shared model powers policies for robots, humanoids, cars, drones, and video games.

Multi-agent simulation (Agora-2)

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 framework (PROWL-2)

The first framework, per the company, in which a team of agents and its world model each learn from their own curriculum, continually and within a single training loop.

Multimodal learning (Starchild-1)

Goes beyond world models that learn only from visual observation, toward systems that learn from richer multimodal interaction with the world.

Interactive environment generation

Generates interactive environments where humans and AI can take actions and experience their consequences, supporting learning through experience.

Breadth of application domains

Targets robotics, self-driving, drones, defense, gaming, energy, cyber, and AI training use cases with the same underlying world model.

How It Works

Learn from the world

World models learn from broad world experience, building knowledge of motion, physics, and how interactions with the world shape what happens next.

Learn from experience

The models generate interactive environments where humans and AI can take actions and experience their consequences.

One shared foundation

A single foundation world model powers robots, humanoids, cars, drones, and game-playing agents, rather than separate systems per domain.

Pros & Cons

Pros

A single foundation world model is designed to power many domains — robots, humanoids, cars, drones, and games — rather than requiring separate simulators per use case
Models learn causal, physics-grounded dynamics (physics, motion, cause-and-effect, agent behaviors) rather than visual appearance alone
Multi-agent support (Agora-2) lets humans and AIs interact in the same real-time simulation
PROWL-2 offers a continual learning setup where agents and the world model each follow their own curriculum in one training loop
Team with experience from DeepMind, Waymo, Tesla, Meta, Apple, and xAI; backed by a $310M Series B

Who It's For

Best for

Robotics and autonomous vehicle teams needing simulation environments grounded in learned physics and dynamics
AI research groups working on reinforcement learning and multi-agent systems
Game developers and studios exploring simulation-based worlds and game-playing agents
Organizations in defense, energy, and cyber exploring simulation-driven AI training
Researchers studying how agents learn through interaction and shared experience in simulated worlds

Not ideal for

Teams seeking a plug-and-play consumer app with no AI research or engineering capability

FAQ

What is Odyssey?

Odyssey is an AI lab pioneering general world models: causal, multimodal systems that learn to understand and simulate the world, with the stated goal of revolutionizing fields like robotics, autonomous driving, science, healthcare, education, gaming, defense, and energy.

What products does Odyssey offer?

Odyssey offers a family of research models: Odyssey-3, its most powerful foundation world model for physical and virtual systems; Agora-2, a multi-agent world model for shared real-time simulation; PROWL-2, a reinforcement learning framework where agents and the world model learn continually in a single training loop; and Starchild-1, a model that learns from richer multimodal interaction with the world.

How does a world model learn?

According to Odyssey, a world model learns to predict the most likely future by capturing physics (how objects behave), dynamics (how systems move and change over time), cause and effect (how actions influence outcomes), and agent behaviors (how agents act and pursue goals).

Can I try Odyssey's models?

Odyssey