Odyssey

Odyssey is an AI lab focused on general world models. Its stated goal is to build causal, multimodal systems that let AI learn physical laws, dynamics, causal relationships, and agent behavior, so machines can understand and simulate the real world rather than just predict text. The company reports $310 million in Series B funding, and its team comes from institutions such as DeepMind, Waymo, Tesla, and Meta. Its flagship product, Odyssey 3, is a foundational world model intended for physical and virtual uses including robotics, autonomous driving, drones, and games. The lab has also launched the multi agent world model Agora 2 and the reinforcement learning framework PROWL 2. It emphasizes learning from real world experience and interaction, publishes open source papers, and offers trial access, with pricing described as mainly free trials and no paid details disclosed.

- Read open source papers describing the world model research

- Request trial access to evaluate the models directly

- Study Odyssey 3 for robotics, driving, drone, or game style use cases

- Experiment with Agora 2 for multi agent settings and PROWL 2 for reinforcement learning

This is aimed at researchers and engineers rather than general users. AI research teams studying world models, causal learning, or reinforcement learning are the clearest fit. Robotics and autonomous driving groups that need models with a sense of physics and interaction will find the positioning relevant. Companies and labs tracking the state of interactive simulation can use the published papers to compare approaches. Anyone evaluating next generation simulation for games or virtual environments may also find something useful.

Odyssey is a research lab, so expect papers, demos, and trial access rather than a finished product with a dashboard and a support desk. Paid plans are not disclosed, which makes long term budgeting difficult, and trial access may be limited or granted selectively. Early stage models can change quickly or be withdrawn as research moves, so anything you build on them should tolerate that. Finally, the claims about understanding and simulating the world are research goals, not guarantees, so evaluate results on your own tasks before relying on them.