AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Odyssey-3 is a new generative world model that you can try for free

Collected Oct 11, 2026

Odyssey, a California-based AI company, has released Odyssey-3 as a public research preview, letting anyone try a world model that generates interactive environments from text prompts in real time. Founders Oliver Cameron and Jeff Hawke first showed the model on September 15 with a focus on robotics, autonomous driving and video games; the new release adds public access, more technical detail and benchmark results. Developers can apply for API access, and a free online demo runs on a lighter variant called Odyssey-3 Flash.

In the demo, users describe an environment in text, then choose first-person or third-person perspective, move through the generated world, trigger events and watch the model react live. Under the hood, Odyssey-3 is an autoregressive diffusion transformer that keeps producing new video frames conditioned on previous frames and user actions. Odyssey says the model picks up physical relationships and cause-and-effect from visual observations during training. The training mix included internet videos paired with event descriptions, video game footage paired with the matching keyboard and mouse inputs, and simulated physical interactions. An additional training technique cuts the number of compute steps required, which is what makes real-time generation feasible.

On specifications, the base Odyssey-3 has 14 billion parameters and outputs video at 832 x 480 pixels, according to its official benchmark submission. A larger configuration, Odyssey-3 Pro, supports 1280 x 720 pixels.

The benchmark picture is where things get interesting, and where care is warranted. Odyssey-3 Pro scores 66.1 points on the video-to-video benchmark from Physics-IQ Verified, which tests physical behavior across areas such as fluid mechanics, optics, solid mechanics, magnetism and thermodynamics. Models must continue videos of real-world experiments, and their continuations are compared against what actually happened. That 66.1 figure, however, comes from a single run in which a selection method picked one of eight generated videos per task. The benchmark's rules call for four runs with a standard deviation reported before any record claim, and the reported record does not meet that bar. Without the selection method, Odyssey-3 Pro averaged 63.37 points across four runs. Both numbers appear on the official leaderboard, and both were submitted by Odyssey itself.

On WorldMark, Odyssey-3 ranks first in three of four categories based on Odyssey's own evaluation: First-Person Stylized (77.2), Third-Person Real (79.0) and Third-Person Stylized (76.3). It places third in First-Person Real with 80.6 points. WorldMark looks at how well models follow control instructions, how good the generated imagery looks, and whether simulated worlds stay consistent over time.

The pitch is that one world model can serve many tasks, from moving robotic arms to navigating drones to controlling game characters. For each application, Odyssey pairs the model with a specialized controller that converts its predictions into concrete commands. In tests, an AI built on Odyssey-3 controlled multiple robotic arms, and Odyssey says a few dozen hours of demonstration data was enough for training. The robots could also recover from failed grasps on their own, even though those situations were not part of the training data. For humanoid robots, Odyssey is working with Swiss robotics company Flexion, and controllers Flexion built on Odyssey-3 are said to perform more reliably than comparison models even when conditions change.

Odyssey also built a drone controller trained on simulated flight data. In a virtual indoor environment, the drone could dodge obstacles and fly to specified targets on command, according to the company. On the games side, Odyssey demonstrated an AI autonomously playing GTA V, steering vehicles and fighting enemies. A controller trained on roughly two hours of GTA footage transferred its skills to Red Dead Redemption 2 without extra training, moving a character on horseback.

A further ambition is agent training. In one demo, an AI agent received a task in natural language and tried to complete it through its own actions inside an Odyssey-3-generated environment. The stated goal is for agents to learn from the outcomes of their own actions, which would be a notably different training paradigm from the current mix of supervised fine-tuning and reinforcement learning against hand-built reward signals.

Odyssey was founded in 2023 by Cameron and Hawke. In June 2026 it raised $310 million from investors including Amazon and AMD Ventures. The world-model field is getting crowded: Google DeepMind is pursuing a similar approach with Genie 3 for interactive world generation, and World Labs, founded by Fei-Fei Li, is building comparable technology. AMD said in late September that it plans to acquire World Labs for roughly $8.2 billion.

Why it matters: A free, interactive preview lowers the barrier for developers who want to test whether generated worlds are good enough for simulation, game prototyping or robot policy training, and the API application path suggests commercial access is coming. The benchmark caveats are the practical caution: self-submitted results and a record that does not follow the stated four-run protocol mean teams should validate physics fidelity on their own tasks rather than relying on leaderboard position. For robotics and product groups, the more interesting signal is the controller-transfer result, since reusing one world model across arms, drones and games is exactly the kind of generalization that is still rare in practice.

Read at The Decoder

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Odyssey is making its world model Odyssey-3 available as a public research preview. The 14-billion-parameter model generates interactive environments from text prompts in real time and can reportedly control robots, drones, and even GTA V. Odyssey claims top scores on physics benchmarks, though the methodology behind those results has limits. The article Odyssey-3 is a new generative world model that you can try for free appeared first on The Decoder .