Odyssey Launches World Model Odyssey-2 Max, Leading in Physics Simulation Scores

According to monitoring by Dongcha Beating, the world model company Odyssey has released Odyssey-2 Max, which has three times the parameter count of its predecessor, Odyssey-2 Pro, and ten times the training computation, making it the company’s largest general-purpose world model to date. Unlike bidirectional video models such as Sora and Veo, Odyssey-2 Max employs a causal autoregressive architecture that predicts the next state frame by frame and accepts real-time interactive inputs, enabling it to continuously generate simulated visuals for over 120 seconds without crashing. In terms of physics simulation accuracy, Odyssey-2 Max scored 58.52 in the VBench 2 physics sub-item, surpassing Odyssey-2 Pro’s score of 49.67 and NVIDIA’s Cosmos-Predict2.5-14B’s score of 44.92; it achieved a score of 93.02 in the PAI-Bench physics sub-item. The model is based on the autoregressive diffusion Transformer (AR DiT) architecture, utilizing continuous flow matching generation and achieving real-time inference through few-step denoising distillation. Training was completed on hundreds of NVIDIA B200 GPUs and was conducted in three phases: large-scale video pre-training, interaction and task conditioning, and long-sequence stability training. Odyssey was founded by Oliver Cameron (CEO) and Jeff Hawke (CTO), both from the autonomous driving industry, and has raised a total of $27 million in funding. Odyssey-2 Max is now available for Private Beta to partners in the fields of robotics, gaming, simulation, defense, and interactive systems.