Odyssey is an AI lab researching “world models,” generating interactive video from text or image prompts that responds to the viewer’s input in real time. What you get is not a clip you simply play back, but a simulation whose footage changes on the spot as you act through a keyboard or controller. Its flagship model, Odyssey-2, was announced in October 2025, and in January 2026 the more capable Odyssey-2 Pro shipped alongside a developer API. In June 2026 the company raised a $310 million Series B at a $1.45 billion valuation.
Key Features
- Interactive video that reacts to your input: Act on the generated footage with a keyboard or controller and the next frame is generated to reflect that input. Frames are produced causally and autoregressively at 20 fps (roughly 50 milliseconds per frame)
- Worlds from either text or images: Describe a situation in words or hand it an image, and the model produces footage that moves as a continuous space. Having learned motion that stays physically consistent is what separates it from one-shot video generation models
- Streams that run for minutes: Rather than stitching short clips together, it generates a stream that continues for minutes while you interact with it. Startup is fast, with the stream beginning shortly after the prompt is submitted
- Multiplayer through Agora-1: Several participants — human or AI agents — can enter the same simulation at once and operate within a shared world
- Related models for research: Starchild-1 is a multimodal model that learns from inputs beyond visual observation, and PROWL-1 is a framework in which reinforcement learning agents explore game environments to strengthen the world model
- Developer API: The developer portal and API documentation arrived with the January 2026 release of Odyssey-2 Pro, making it possible to call simulations from your own application
Pricing
| Plan | Price | What it covers |
|---|---|---|
| Research preview (experience.odyssey.ml) | Free | Try interactive simulations with Odyssey-2 and Agora-1 straight from the browser |
| Developer API | No public price list published (check directly) | Sign-up through the developer portal. At launch, a pool of one million free simulations was offered first-come, first-served, with up to 10 concurrent runs per user |
Pricing is current as of August 2026. No public pricing page for API usage could be found, so please confirm the latest terms on the official site and the developer portal.
Pros & Cons
✅ Pros
- Unlike generated video you only watch, you act and get a result back, so you can build the experience itself
- A free research preview runs in the browser, reachable without knowledge of 3D modeling or game engines
- Because output arrives as a minutes-long stream, there is no editing work to splice short clips together
- Multiplayer lets several people enter the same world, which makes collaborative experimentation easy
- With the API published, the path from experiment to embedding it in your own product is open
⚠️ Cons
- The technology is still at a research stage, and the consistency and fine detail of the generated world vary by scene
- It is not suited to exporting what you generated as 3D assets or game data — the output is a video stream
- Public information on API pricing is thin, making costs hard to estimate before a serious rollout
- Real-time generation is sensitive to network conditions, so comfortable operation depends on the environment
- Official documentation and commentary in Japanese are nearly nonexistent for now, so following English sources is required
Comparison with Similar Services
| Item | Odyssey | Google Genie 3 | Decart | World Labs |
|---|---|---|---|---|
| Provider | Odyssey | Google DeepMind | Decart | World Labs |
| What is generated | An interactive video stream that reacts to input | Interactive environments you can operate | Real-time video transformation and interactive worlds | 3D spaces you can explore |
| How you interact | Keyboard, controller, and similar | Keyboard and similar | Live video input and key operation | Exploration by moving the viewpoint |
| Simultaneous multi-user | Supported via Agora-1 | Not indicated in public information | Not indicated in public information | Not indicated in public information |
| General availability | Research preview plus developer API | Mostly limited availability | Product and API available | Product available |
Who Is It For
- Anyone who wants to touch “video you can operate,” the step after video generation, and see it firsthand
- Creators who want to quickly test the feel of a setting during the planning stage of a game or interactive piece
- Planners and designers who want to stand up an experience prototype without preparing 3D assets
- Developers who want to treat world models and simulated environments as material for AI research or reinforcement learning
- Developers exploring whether to embed interactive simulation into their own application
Summary
Odyssey is an AI lab trying to move the output of generative AI from something you watch to something you operate. Odyssey-2 delivered interactive streams lasting minutes, Agora-1 added simultaneous participation by several people, Starchild-1 and PROWL-1 widened the reach into research, and an API arrived in 2026. Some roughness from its research stage remains, but because it is free to try in the browser, it is a good way to confirm with your own hands how far world models have come.