# Odyssey-3

Odyssey-3 is a learned simulator: an autoregressive diffusion transformer that builds an environment from a prompt and predicts how it changes as a person or an agent acts. Odyssey aims it at physical-AI work and offers it as a research preview.

Organization: Odyssey. Released: 8 October 2026. Kind: Physical AI.

## What you control

- A prompt that creates the environment
- Actions from a person or an AI agent

## License

Weights: Not published in cited source. Declared license: No public weight license verified.
No weights are published, so there is no model license to use. Commercial use depends on the provider's own service terms.

## Real-time

Real-time, per authors. Predicts how the environment changes in real time; no frame rate published.

## Where to try

- [Official](https://experience.odyssey.systems/)

## Sources

- [Official source](https://odyssey.systems/meet-odyssey-3)

## Questions

### Is Odyssey-3 open source?

No. Odyssey-3 has no public weights; it is available only as a hosted service or demo.

### Can I use Odyssey-3 commercially?

No weights are published, so there is no model license to use. Commercial use depends on the provider's own service terms.

### Does Odyssey-3 run in real time?

Yes, according to its authors. Predicts how the environment changes in real time; no frame rate published. This directory has not measured it independently.

### Where can I try Odyssey-3?

You can try it on the official site.

### What does Odyssey-3 take as input?

You give it a prompt that creates the environment and actions from a person or an AI agent.

### What are alternatives to Odyssey-3?

Similar models in this directory include Cosmos Predict2.5 (open weights, custom terms), Agora (Odyssey) (no public weights, service terms) and Genie 3 (no public weights, service terms).

Checked 2026-10-23. Corrections: https://github.com/windflow-labs/world-model-directory