What is Odyssey? World Model Odyssey-3 Guide Home

HomeOliver Cameron and Jeff Hawke, who worked for years in the autonomous vehicle field, is an artificial intelligence laboratory founded in 2023. What it develops is not a chat bot or a great language model (LLM) world model (world model): a model that learns physics, movement and cause-end relationship from videos, predicting the next moment of an environment, producing real-time and interactive simulation. The most current model of the company opened as research preview at 08.10.2026 Odyssey-3.

Last update: 14.10.2026

HomeThe first showed a model that converts a single image into a detailed 3D world with Explorer introduced in 18.12.2024. Today, the same approach is trying to move to a common basic model for robots, tools, drones and games.

What is Odyssey Artificial Intelligence Model?

Briefly: Odyssey is not a language model that produces text; a visual observations that process the world world model. The world model with its own definition of the company is a dynamic model connected to the action in essence: the current state takes the action and history, the next situation estimates as a video square. That’s why the image will vary accordingly when you are in progress in one direction or write a script.

Text production, translation or code writing is not the work of Odyssey. Odyssey's models produce interactive video and 3D environment; the latest version is Odyssey-3, with the same model, it is aimed to control the robot, vehicle, drone and gaming agents.

Odyssey’s first product that everyone can try is opened as research preview at 28.05.2025 Odyssey-1 It was: real-time, playable interactive video experience. From 27.10.2025 Odyssey-2 It produces an interactive video stream that reacts instantly as soon as written. The company identified these models as "the stage" from the beginning.

For example, Odyssey artificial intelligence model created video:

https://www.youtube.com/watch?v=xq KMs-Dbng

The sample in the video belongs to the early period. Odyssey said that Odyssey-1 experience itself was unstable as raw and unstable as "to navigate in a dream." The next versions are concentrated on physical accuracy and consistency.

Odyssey models and dates Home

Tags

18.12.2024

The first world model that converts a single image into a detailed 3D world

Odyssey-1

28.05.2025

Real-time, playable interactive video (research preview)

Odyssey-2

27.10.2025

Interactive video that reacts instantly to written commands, lasts minutes

Odyssey-2 Max

21.04.2026

Physical accuracy augmented, scaled world simulation u

Starchild-1

18.05.2026

Model "first real-time multi-mode world model"

Agora-1 / Agora-2

18.05.2026 / 24. 09.2026

The world model with multiple people or artificial intelligence to real-time impact in the same simulated world

PROWL-1 / PROWL-2

Tel.: +55 11 311 381

Consolidation learning frame that agents and world models learn together

Odyssey-3

15.09.2026 (push), 08.10.2026 (removing)

Common basic world model for robot, vehicle, drone and games

What Does Odyssey Like Tech?

Technical specifications that stand out according to Odyssey's official descriptions are:

• Odyssey-3, an autoregresific diffusion transformer: produces each new square looking at the previous squares and the actions of the user or the agent.

• From a wide range of visual observation data sets, physics, dynamic, cause-gain and human behaviors learn.

• With Starchild-1, it’s not only from the image, but it’s going to learn from a richer mode interaction.

• Odyssey-1 was able to produce a new square in each 40 milliseconds and give 5 minutes and more consistent video stream.

• Odyssey-3 was able to learn to control different robot arms with dozens of hours of robot display in company’s experiments; Odyssey-3 The Pro scored 66,1 in the format of Physics-IQ Verified (the result that the company announced).

Uses of Odyssey

Odyssey is positioned for physical and virtual systems, not for chat or text works. The main areas of use of the company:

• Robotics: Learning the ability to control robot handles and human robots with less data. For human robots this compilation You can browse.

• Autonomous driving and drone: Directing systems such as vehicle and drone with the same basic model.

• Artificial Intelligence Training Home: Producing simulated environments where agents can learn by experimenting (PROWL-2).

• Game, movies and interactive media: 3D world (Explorer) and interactive video (Odyssey-2).

• Multi-existing simulation: People and artificial intelligence agents with Agora-2 can influence real-time in the same simulated world.

Also, characters created with artificial intelligence You can also browse.

Where Will Odyssey In The Future?

Odyssey’s own goal is to be common foundation for many machines and tasks of a single world model. The company thinks that the "GPT-3 moment" for world models is approached; but it will show you how much it will hold. The following directions of the company are:

1. Robot training with less data: The company’s goal is to use the general knowledge of the world model instead of collecting large amounts of data for each task and teach new skills to robots with several hours or dozens of hours of experience.

2. Simulation trial: Developers can use Odyssey-3 to simulate both environment and train control policy for different physical systems. Thus, a system can be tried in simulated environment without leaving the real world.

3. World production for game and film: Odyssey said that 3D worlds are a big bottleneck in the film and game to produce manually while promoting Explorer, can speed up this process of world models. Pixar’s co-founder Ed Catmull also joined the company’s board with this announcement.

4. Educational field for artificial intelligence agents: With Agora-2 and PROWL-2, the company works on learning by experimenting artificial intelligence agents in the shared simulated worlds.

1. Open questions: Odyssey also researches that the model does not show the same success in every robot and environment; for this, it tests different robots and controls with Poke & Wiggle. So the results are promising but at the early stage.

2. Access: Odyssey-3 can now be said on the web as research preview; for developers, API access is provided with reference. Pricing information is not open on the official site.

Odyssey announced that it invested in 310 million dollar B series over $1,45 billion in 17.06.2026; tour managed Natural Capital, Amazon, GV, AMD Ventures, EQT and IQT also joined. Amazon Web Services was the cloud provider preferred by the company. This shows that world models are a big investment area, but Odyssey himself says "We are at the earliest days we are on.

If you want to follow the developments in artificial intelligence Masqot communities join, events You can browse.

Frequently Asked Questions

Is Odyssey a chat boat like ChatGPT?

No. Odyssey is not a language model, the world model is developing: models that predict the next moment of a media instead of text and generate interactive video or simulation.

What is the world model?

The world model is an artificial intelligence model that will learn the physics, movement and cause-end relationship and predict how the environment will change after an action. Odyssey produces this prediction as video frames.

Which is the newest model of Odyssey?

14.10.2026 is the newest model Odyssey-3. Published in 15.09.2026, the research preview was opened in 08.10.2026.

How can I try Odyssey?

Odyssey-3’s research preview can be said from the “Try Odyssey-3” link on the company’s site; applications from the developer page for API access.

Who created Odyssey?

Hedyssey founded Oliver Cameron and Jeff Hawke, working on autonomous vehicles in 2023. Jeff Hawke’s technology director (CTO).