Odyssey-3: Free World Model Preview & Access

Odyssey-3 is Odyssey’s interactive world model: type a prompt and explore a world it generates in real time. See how to try it free, pricing and limits.

Odyssey-3 is one of the AI world models people are sharing clips from this month. Odyssey first showed it on September 15, 2026, then opened a public research preview on October 8. This page explains what Odyssey-3 is, who Odyssey AI is, how to try the free preview in your browser, what Pro and API access look like and where the model still falls short. Everything was checked on Odyssey’s site on October 20, 2026.

What is Odyssey-3?

Odyssey-3 is a foundation world model: an AI that predicts how a scene changes over time as you act in it. In the launch post, Odyssey describes it as a learned dynamical system, built as an autoregressive diffusion transformer, that predicts how objects move and interact and how situations unfold. In plain terms, you describe a world, it starts generating video, and it keeps generating in real time as you move or trigger events.

• Real-time interaction: first-person and third-person navigation, plus independent camera movement.
• Events: you can introduce an event mid-generation and see how the world responds.
• Two sizes: the base Odyssey-3 generates at 832×480 and Odyssey-3 Pro at 1280×720. The free preview runs Odyssey-3 Flash. Odyssey says it distills the model into a few-step variant for real-time interaction.
• Physics focus: Odyssey trained it on internet video with event annotations, gameplay with keyboard and mouse inputs, and simulated physical interactions.

How good is it?

Odyssey says Odyssey-3 Pro set a new high of 66.1 on the Physics-IQ Verified video-to-video benchmark and ranks first in three of WorldMark’s four categories in its own evaluation. There’s a catch worth knowing: as The Decoder reported, the 66.1 comes from a single best-of-8 run, and without that selection Pro averaged 63.37 over four runs. Both results were submitted by Odyssey itself.

What is Odyssey AI?

Odyssey is a California AI company founded in 2023 by Oliver Cameron and Jeff Hawke, whose team spent a decade on driverless cars. Its site says it raised a $310M Series B. The goal is bigger than game-like demos: Odyssey wants one world model to power robots, humanoids, cars, drones and AI agents. In its tests, Odyssey-3 controlled robot arms after tens of hours of demonstrations, powered humanoid policies built by Flexion, and drove a car on real roads in India after training on 20 hours of driving data.

Note that other companies use the Odyssey name. The world-model company lives at odyssey.systems.

Is Odyssey-3 free?

Yes, the research preview is free to try in your browser. It runs Odyssey-3 Flash: you type a prompt, an environment appears, and you can move through it and trigger events. You don’t need a GPU, because it runs on Odyssey’s servers.

Keep expectations in check: it’s a research preview, not a product with guaranteed access. When we opened it on October 20, a countdown timer was running in the corner, so sessions appear to be time-limited.

Odyssey-3 Pricing (2026)

Odyssey hasn’t published any prices. Here’s what access looks like on October 20, 2026:

Access	Price	How to get it
Odyssey-3 Flash research preview	Free	Try Odyssey-3 at odyssey.systems in your browser
Odyssey-3 Pro	Not public	Get in touch with Odyssey
API for developers	Not public	Request access as a physical AI developer
Open weights	Not available	Odyssey hasn’t released the model

How to try Odyssey-3

1. Open the preview: click Try Odyssey-3 on odyssey.systems, which takes you to the Odyssey experience.
2. Describe a world: type a scene in the prompt box, or pick one of the sample worlds. You can also start from an image.
3. Choose a view: switch between first-person and third-person.
4. Move and act: walk around, move the camera and introduce events, then watch how the model predicts what happens next.

Our tip: simple, physical prompts work best for testing a world model. Try something like water pouring, a ball rolling down a street or an animal walking through grass, then push or change something and see whether the world reacts the way physics says it should.

Limitations

• Research preview: access and session length can change, and quality varies.
• Low resolution: 832×480 for the base model. The 720p Pro model isn’t in the free preview.
• No public pricing or API: developers have to request access.
• Closed model: no open weights, so you can’t run it on your own GPU.
• Self-reported benchmarks: the Physics-IQ and WorldMark results come from Odyssey’s own submissions and evaluation.

Odyssey-3 alternatives

• HelixWorld: Noiz AI’s free, open-source world model that adds stereo sound. See our HelixWorld page.
• Google DeepMind Genie 3: a similar interactive world model from Google.
• If you want a real game: world models are still research demos. See how to make a game with AI for tools that build playable games.
• If you only need clips: a normal video model is simpler. Start with our free AI text to video generators.

Verdict

Odyssey-3 is worth five minutes of your time: the free browser preview is the easiest way to feel what a real-time world model can do, and Odyssey’s robotics and driving results show why investors are betting on it. For anything beyond play, though, you’ll need to contact Odyssey, because there’s no public price, API or open model yet.

Odyssey-3 is an impressive, free way to try a real-time AI world model in your browser, backed by serious physics and robotics research.

It’s still a research preview, with no public pricing, API or open weights, so treat it as a demo rather than a production tool.