ChainThink report, April 22: According to an official announcement, World Model company Odyssey has launched Odyssey-2 Max.

The model has three times the parameters of its predecessor Odyssey-2 Pro and ten times the training compute, making it the largest general-purpose world model developed by the company to date.

Differing from bidirectional video models such as Sora and Veo, Odyssey-2 Max adopts a causal autoregressive architecture, predicting the next state frame-by-frame while accepting real-time interaction input, enabling continuous generation of simulated scenes for over 120 seconds without collapse.

In terms of physical simulation accuracy, Odyssey-2 Max scores 58.52 on the VBench 2 physics sub-benchmark—surpassing Odyssey-2 Pro’s 49.67 and NVIDIA Cosmos-Predict2.5-14B’s 44.92;

and achieves a score of 93.02 on the PAI-Bench physics sub-benchmark.

The model is built upon an autoregressive diffusion Transformer (AR DiT) architecture, generating outputs via continuous flow matching and enabling real-time inference through few-step denoising distillation. Training was completed in three stages across hundreds of NVIDIA B200 GPUs, including large-scale video pre-training, interaction and task conditioning, and long-sequence stability training.

Odyssey was founded by CEO Oliver Cameron and CTO Jeff Hawke, both with backgrounds in the autonomous driving industry, and has raised a total of $27 million in funding to date.

Odyssey-2 Max is now available in Private Beta to partners in robotics, gaming, simulation, defense, and interactive systems.