Odyssey

Odyssey is an AI research lab focused on building general-purpose world models — causal, multimodal systems that learn to predict and interact with the world over long time horizons. The company is headquartered in Palo Alto, CA with a presence in London, GB, and is backed by investors including Elad Gil, Jeff Dean, Guillermo Rauch, and Garry Tan.

What It Is

Odyssey describes its mission as pioneering world models that go "beyond language models." Rather than generating fixed video clips, Odyssey's models produce interactive, streaming simulations that users can prompt and steer in real time. The lab positions world models as a foundational technology for robotics, gaming, healthcare navigation, retail training, and other domains where AI must understand and act within physical environments.

The Model Lineup

Odyssey has released several distinct models, each advancing a different dimension of world modeling:

• Odyssey-2 / Odyssey-2 Max — Described as the lab's most powerful general-purpose world model, it begins streaming an interactive simulation in approximately 50ms and can sustain multi-minute sessions. Unlike traditional video generation models that take minutes to produce a fixed 10-second clip, Odyssey-2 streams frame-by-frame and supports open-ended interaction.
• Starchild-1 — Billed as the first real-time multimodal world model, Starchild-1 goes beyond visual observation to learn from richer multimodal interaction. A technical report is publicly available.
• Agora-1 — A multi-agent world model that allows multiple participants — human or AI — to share and interact within the same simulated world simultaneously in real time.
• PROWL — A reinforcement learning framework where an RL agent adversarially explores game environments to improve world model performance. A research paper is available.

How the Simulation Works

Odyssey-2 distinguishes itself from conventional video generation by streaming rather than rendering. The homepage states it "began simulating in 50ms" and supports "minutes of streaming simulation." Users prompt the model with an image or text, and the model rolls out an interactive simulation rather than a pre-rendered clip. This architecture is designed to support open-ended interaction, making it suitable for robotics rehearsal, game environments, and real-time applications.

Intended Applications

Odyssey's site lists a broad range of envisioned use cases, including:

• Accelerated robotic intelligence and edge-case scenario generation for autonomous systems
• Reinvention of gaming through AI-generated interactive worlds
• Intelligent guides for hospitality, healthcare navigation, and customer support
• Personalized operating systems and interactive fitness software
• Live, immersive events and interactive advertising

Team and Backing

The leadership team includes Co-Founder & CEO Oliver Cameron, Co-Founder & CTO Jeff Hawke, VP of Engineering James Grieve, VP of GTM & Operations Jessica Inman, and Distinguished Engineer Fabian Güra. The About page states the team is recruited from organizations including leading AI research institutions and technology companies. The company is backed by prominent angel investors and venture firms.

Update: Agora-1 Launch

The most recent product announcement is Agora-1, the multi-agent world model that enables multiple participants to share a real-time simulation. Prior releases include Starchild-1 (the first real-time multimodal world model) and Odyssey-2 Max (scaled world simulation). The active blog and research output signal ongoing model development, with the lab publicly framing world models as a nascent but rapidly advancing technology category.