Odyssey opens Odyssey-3 world model to the public with a free demo

California-based AI company Odyssey has launched a public research preview of its world model Odyssey-3. Users can generate interactive environments from text prompts and explore them in real time. The model simulates physical processes and predicts how environments change based on specific actions. The free online demo runs on Odyssey-3 Flash, and developers can apply for API access. Founders Oliver Cameron and Jeff Hawke first unveiled the model on September 15, with a focus on robotics, autonomous driving and video games. What is new now is public access, more technical detail and benchmark results.

In the demo, users choose between first-person and third-person perspectives, move through the generated world, trigger events and watch the model respond. Odyssey-3 is built on an autoregressive diffusion transformer that continuously generates new video frames from previous frames and user actions. According to Odyssey, it learns physical relationships and cause-and-effect from visual observations during training. Training data included internet videos with event descriptions, video game footage paired with keyboard and mouse inputs, and simulated physical interactions. An extra training technique reduces the number of required compute steps, which makes real-time generation possible. The base model has 14 billion parameters and generates video at 832 × 480 pixels, per the official benchmark submission; Odyssey-3 Pro supports 1280 × 720 pixels.

On benchmarks, the company says Odyssey-3 Pro scores 66.1 points on the video-to-video benchmark from Physics-IQ Verified, which tests physical behavior across fluid mechanics, optics, solid mechanics, magnetism and thermodynamics by having models continue videos of real-world experiments. That score comes from a single run in which a selection method picked one of eight generated videos for each task. The benchmark rules require four test runs with standard deviation reported for any record claim, so the reported record does not meet that bar. Without the selection method, Odyssey-3 Pro averaged 63.37 points across four runs. Both results appear on the official leaderboard but were submitted by Odyssey itself. On WorldMark, which tests control-instruction following, visual quality and consistency over time, Odyssey-3 ranks first in three of four categories based on its own evaluation: First-Person Stylized (77.2), Third-Person Real (79.0) and Third-Person Stylized (76.3). In First-Person Real it places third with 80.6 points.

Odyssey wants one world model to serve different tasks, each paired with a specialized controller that translates the model's predictions into commands. According to the company, an AI built on Odyssey-3 controlled multiple robotic arms after a few dozen hours of demonstration data, and could recover from failed grasps on its own even though those situations were not in the training data. For humanoid robots Odyssey works with Swiss robotics company Flexion; controllers Flexion built on Odyssey-3 are said to perform more reliably than comparison models, even when conditions change. A drone controller trained on simulated flight data could dodge obstacles and fly to specific targets in a virtual indoor environment. In games, the company showed an AI playing GTA V, steering vehicles and fighting enemies; a controller trained on roughly two hours of GTA footage transferred to Red Dead Redemption 2 without extra training, moving a character on horseback.

Odyssey also plans to use the model to train AI agents. In one demo an agent received a natural-language task and tried to complete it through its own actions inside an Odyssey-3-generated environment. The goal is for agents to learn from the outcomes of their own actions, a potentially new training paradigm. For now the public preview offers interactive environment generation; robotics and autonomous systems need further work.

Odyssey was founded in 2023 by Oliver Cameron and Jeff Hawke, and in June 2026 raised $310 million from investors including Amazon and AMD Ventures. Google DeepMind pursues a similar approach with Genie 3, and World Labs, the startup founded by Fei-Fei Li, is developing similar technology; AMD announced in late September that it plans to acquire World Labs for roughly $8.2 billion.

Key facts

Odyssey launched a public research preview of Odyssey-3, a world model that generates interactive environments from text prompts in real time; the free demo runs on Odyssey-3 Flash and developers can apply for API access.
The base model has 14 billion parameters and outputs 832 × 480 video; Odyssey-3 Pro supports 1280 × 720.
Odyssey-3 Pro's 66.1 on Physics-IQ Verified comes from a single run with a selection method picking one of eight videos; without it, the four-run average is 63.37, and the rules require four runs for a record claim.
On WorldMark, Odyssey-3 ranks first in three of four categories by Odyssey's own evaluation (77.2, 79.0, 76.3) and third in First-Person Real with 80.6.
The company shows one model driving robotic arms, a drone and game characters via specialized controllers, including a GTA-trained controller that transferred to Red Dead Redemption 2.

Why it matters

World models aim to give AI systems an understanding of spatial and physical relationships, and Odyssey-3 moves from an announcement on September 15 to something anyone can try. The company's pitch is a single model that can be paired with controllers for robotic arms, drones and game characters, and later used to let agents learn from the outcomes of their own actions, which the article calls a potentially new training paradigm. It lands in a crowded field: Google DeepMind has Genie 3, and World Labs, founded by Fei-Fei Li, is developing similar technology.

Who it affects

Developers can apply for API access, so the first audience is people building on interactive world generation. The robotics, drone and game demos are aimed at teams working on robot control, autonomous systems and game AI, with Flexion already building humanoid-robot controllers on the model. Competing world-model groups, including Google DeepMind and World Labs, are the reference points the article draws.

How to use it

Open the free online demo, which runs on Odyssey-3 Flash. Describe an environment in text, pick a first-person or third-person view, move through the world, trigger events and watch the model respond in real time. Developers who want programmatic access can apply for the API. The source gives no pricing, usage limits or duration of free access, and it does not say whether Odyssey-3 Pro is available in the public demo.

How solid is it

Mixed. The launch and demo are facts; most performance claims are the company's own. The Physics-IQ record of 66.1 does not meet the benchmark's four-run requirement, and the four-run average without the selection method is 63.37. Both results are on the official leaderboard but were submitted by Odyssey. The WorldMark rankings rest on Odyssey's own evaluation. The robotics, drone and game results are described as shown or reported by the company.

Risks and caveats

This is a research preview, and the article says robotics and autonomous systems will need further work; no timescale is given. Flexion's reliability claim is worded as "said to" and does not name the comparison models. The headline Physics-IQ number depends on a selection step that picks the best of eight generated videos, so it should not be read as a typical single-pass result. The source gives no frame rate or latency figures for the real-time generation.

“The reported record doesn't meet that bar.”

— The Decoder, on Odyssey-3 Pro's Physics-IQ result