VAST AI Research presents Project Eden, a persistent and shared world model

VAST AI Research unveils Project Eden, a persistent, shared 3D world model funded by Tripo AI's $200 million raise to enable real-time multi-user simulation.

VAST AI Research, the team behind the 3D generation platform Tripo AI, introduces Project Eden, a research preview still far from being a product. Its unique feature: Eden separates the world state from its visual rendering. Where traditional video models produce a sequence of images that do not persist and lose their coherence as soon as the viewpoint changes, Eden treats the world as a structured and persistent environment.

This environment runs continuously, including off-screen, can be modified by user actions or AI agents, and remains consistent regardless of the observation angle. Furthermore, it is shared in real time, with multiple users or agents evolving within the same world with the same physical reality. The system is based on three layers: a structured and evolving camera-independent state, an interface that translates this state into what each individual perceives, and a neural renderer that generates images in real time.

Demonstrations illustrate this logic: a fire remains extinguished when one returns or changes angle, and the same race is simultaneously followed from multiple viewpoints. The team sees this as an exploration of world models useful for autonomous agents, interactive 3D creation, and simulation. Tripo AI, which has achieved unicorn status, reports having raised nearly $200 million to accelerate this work.