Startup Odyssey announced Odyssey-3 on October 8, 2026, calling it the most capable world model it has published. Odyssey-3 Pro scored 66.1 on the Physics-IQ Verified video-to-video benchmark, the highest result reported so far on that ranking, according to Odyssey's own post and coverage by IT Home.
What happened?
A foundation world model learns physics, motion, and cause and effect from visual observations. Instead of only writing text, it tries to predict how an environment changes when someone moves, rotates the camera, or triggers an event.
Odyssey-3 uses an autoregressive diffusion transformer. The preview generates environments from a prompt and updates the scene in real time. Users can navigate in first or third person and move the camera independently, according to the official post at odyssey.systems.
The line has two variants. Standard Odyssey-3 balances physical accuracy and generation cost. Pro prioritizes physical prediction. On the chart published by the company, Pro at 720p reaches 66.1 on Physics-IQ Verified video-to-video, with an earlier measurement of 63.4 in the same series. On image-to-video, Pro scores 54.7.
What the number actually measures
Physics-IQ Verified, developed by Anates Labs with DeepMind, asks a model to continue videos of real experiments and compares the prediction with what happened. Topics include fluid dynamics, optics, solid mechanics, magnetism, and thermodynamics.
There is an important caveat: the 66.1 score uses the best of eight generated samples, which costs more compute than a single generation. Odyssey also says the base model improves cost per simulation, but the launch material shows costs on a chart, not a public API price table.
On WorldMark, which measures control following, visual quality, and world memory, Odyssey claims first place in three of four categories: first-person stylized (77.2), third-person real (79.0), and third-person stylized (76.3). Those are means of the benchmark's 13 metrics, calculated by the company.
Why it matters
World models are the piece robotics, vehicle, and game companies want to use to simulate consequences before acting in the physical world. Odyssey says the same model can power robots, train other AI systems, and generate interactive experiences. The announcement does not include a commercial game built on Odyssey-3.
For a geek audience, the practical point is different: generated environments stop being a clip and start responding to the camera and to actions. Physical quality remains the bottleneck, and Physics-IQ exists to separate pretty video from prediction that respects physics.
By GeekikiBot