Odyssey releases a research preview of the Odyssey-3 model with a free interactive demo
Odyssey has opened a public research preview of the Odyssey-3 model. The free online demo uses the Odyssey-3 Flash model and lets users create environments from text prompts, explore them and trigger events in real time. Developers can request API access.
Odyssey has made a public research preview of the Odyssey-3 model available for generating interactive environments. The free online demo runs on the Odyssey-3 Flash model. Users enter text descriptions, choose a first-person or third-person view, move through the environment and trigger events that the model responds to in real time. Developers can request API access.
The Odyssey-3 model continuously generates subsequent video frames based on previous frames and user actions. According to Odyssey, it learns physical relationships and cause and effect from visual observations during training. According to the official benchmark submission, the base Odyssey-3 model has 14 billion parameters and generates video at a resolution of 832 × 480 pixels. The Odyssey-3 Pro model supports 1280 × 720 pixels; these figures do not describe the specifications of the free demo.
Odyssey reports a score of 66.1 points for the Odyssey-3 Pro model in the Physics-IQ Verified test. However, the result comes from a single run in which one of eight generated videos was selected for each task, so it does not meet the rules for recognition as a record. Without this selection, the model averaged 63.37 points across four runs. Odyssey itself submitted both results to the leaderboard. The public preview currently offers interactive environment creation; use in robotics and autonomous systems requires further work.
Why it matters
The free demo lets users directly assess how the generated environment responds to controls and whether it remains consistent as they move through it. For development teams, requesting API access opens a path to experiments with their own integrations. However, the physics test results alone do not confirm reliability when controlling real devices.
Two audiences, two different impacts
What this means
For individuals
When trying out an idea for a virtual scene, you can observe how it responds to your own movement and events, rather than just evaluating a video generated in advance.
For a business
Corporate development teams can request API access for experimental integrations of generated environments. The available research preview does not yet constitute a ready-to-use solution for controlling robots or autonomous systems.
DevelopmentCheck the original
Event sources
only one source so far · 1 publisher, 1 independent. We count feeds from the same owner only once.