← All companies

World models / Company profile

Odyssey

What Odyssey does

Odyssey is an AI lab building “world models”—systems designed to understand and simulate the world over time, including in response to actions—rather than only generating fixed clips from prompts. The company positions its technology as a foundation for interactive, real-time simulation across domains such as robotics, gaming, science, healthcare, education, and defense.

On the product side, Odyssey has publicly previewed multiple model families that expand the scope of simulation: Odyssey-2 (general-purpose world simulation), Odyssey-2 Max (a larger/scaled variant described as advancing physics accuracy and able to run in real time), Starchild-1 (a “multimodal” world model jointly generating audio-video in real time with streaming user input), and Agora-1 (a multi-agent world model designed to support multiple participants sharing one generated world state in real time).

Odyssey also supports creator workflows through “Explorer,” which it describes as an image-to-world model that transforms images into realized 3D worlds, with outputs described as editable in common 3D/VFX tooling and with a cited collaboration deploying Explorer-generated worlds in a virtual production stage workflow.

Business model details are not fully disclosed publicly, but Odyssey’s site emphasizes access to models via API and “private beta” for larger models, plus ongoing “partner” collaboration for robotics, gaming, simulation, defense, and interactive systems.

News

Company record

Jul 08, 2026 · Official · OdysseyOur Path to Superintelligence

Oliver Cameron posted Odyssey’s argument for open-ended learning shaped by dynamic environments, framing world models as a route to higher ceilings than static, hand-authored training environments.

Jun 17, 2026 · Reporting · TechCrunchOdyssey nabs $1.45B valuation backed by Amazon and other big names (Series B close)

Odyssey raised a $310M Series B at a $1.45B valuation led by Natural Capital, with participation including Amazon, AMD Ventures, GV, and others; TechCrunch also identifies the company’s founders as CEO Oliver Cameron and CTO Jeff Hawke.

Jun 12, 2026 · Official · OdysseyThe Era of Multi-Agent Imagined Experience

Odyssey published a research/product perspective on multi-agent imagined experience, building on its world model work including Agora-1 and PROWL-1.

May 18, 2026 · Official · OdysseyAgora-1: The Multi-Agent World Model

Odyssey released Agora-1, described as enabling multiple participants to share and interact within the same generated world simulation in real time, and positioning it as a learned game engine.

May 17, 2026 · Official · OdysseyStarchild-1: The First Real-Time Multimodal World Model

Odyssey previewed Starchild-1 as a multimodal world model that jointly generates synchronized audio and video in real time and responds continuously to streaming user input.

May 12, 2026 · Official · OdysseyIntroducing PROWL-1: Learning Through Discovery

Odyssey presented PROWL-1, describing a reinforcement-learning-driven adversarial framework that discovers where a world model fails and turns those failures into an adversarial curriculum for improvement.

Apr 21, 2026 · Official · OdysseyOdyssey-2 Max: Scaled World Simulation

Odyssey introduced Odyssey-2 Max, describing it as its largest general-purpose world model yet, highlighting real-time generation, physics benchmark improvements (including VBench 2 physics), and private beta availability for partner applications.

Feb 12, 2026 · Official · OdysseyOdyssey Announces Investment from NVentures and Samsung Next

Odyssey announced an investment from NVentures and Samsung Next aimed at accelerating research toward a general-purpose world simulator.

Show 1 earlier updates
Source map · 2 recurring channels · 20 references

Still resolving: Newsroom · Official X

Funding

Latest disclosed valuation$1.5B
Tracked capital$337M4 sourced rounds

What it builds

Odyssey-2

A general-purpose world model that can generate “minutes-long interactive simulations” and is positioned as a causal, action-conditioned world simulator.

source ↗
Odyssey-2 Max

Odyssey’s scaled general-purpose world model described as advancing physics accuracy while generating real-time rollouts; the company describes private beta availability to partners working on applications such as robotics, gaming, simulation, defense, and interactive systems.

source ↗
Starchild-1

A multimodal world model described as jointly generating audio-video in real time and responding continuously to streaming user input.

source ↗
Agora-1

A multi-agent world model described as enabling up to four players (human or AI) to interact within the same generated world simulation in real time, functioning like a learned game engine.

source ↗
PROWL-1

A reinforcement-learning-driven adversarial framework Odyssey describes as improving world models by having an RL agent discover failures and build an adversarial curriculum for training.

source ↗
Explorer

An image-to-world model described as transforming images into realized 3D worlds using Gaussian splats; Odyssey describes an emphasis on photorealistic world generation and editability in common 3D/VFX tools.

source ↗

Key metrics

Milestones & partnerships