Odyssey-3 world model trains a driving policy on 20 hours of simulation
Odyssey released Odyssey-3, a foundation world model built on an autoregressive diffusion transformer. A driving policy trained on roughly 20 hours of generated video drove on real roads in India. A free research preview is available without an API key.
- Base tier runs at 832×480 and 16 fps, Pro at 1280×720
- Scored 66.1 on the Physics-IQ Verified video-to-video track
- One pretrained model generalizes to cars, humanoids, drones and games
- Speed comes from adversarial distillation of the diffusion sampler
Read next
AI