Odyssey builds causal, autoregressive world models trained on large-scale video and interaction data. Their Odyssey-2 series uses next-state prediction to learn physics, dynamics, and action-conditioned behavior, generating real-time interactive video streams. Odyssey-2 Max leads on VBench 2 physics scoring. API available; robotics and simulation are core target applications.
Odyssey's new framework pairs agents with a self-repairing world model, lifting StarCraft win rates by up to 91% and cracking the hardest robot coordination tasks.
A new peer-reviewed benchmark from Odyssey shows frontier video models collapse to single outcomes instead of reproducing real-world probability distributions.
Odyssey closes a $310M Series B at a $1.45B valuation, betting that world models — AI that simulates physics and causality — are the next foundation model category after LLMs.