Oasis 3 builds on the foundation of the company’s existing real-time video model, Lucy. Priced at $0.02 per second, the system leverages Decart’s proprietary optimization stack to run efficiently on hardware from Nvidia, Amazon, and Google. CEO Dean Leitersdorf emphasizes that this vertical integration allows the platform to operate at a fraction of the cost typical of current world models. Following a recent $300 million funding round that pushed its valuation toward $4 billion, the company is positioning itself to compete with rivals like Google’s Genie 3 and World Labs’ Marble.
Despite the impressive visual fidelity, the model currently faces significant technical hurdles regarding long-term consistency. In practice, generated environments often lose their initial thematic integrity after extended use, and physics simulations remain imperfect, with vehicles occasionally passing through solid objects. Leitersdorf attributes these artifacts to the auto-regressive nature of the architecture, where the context window fills rapidly, and a lack of training data regarding collision scenarios. The team is now prioritizing research into longer-context memory and video-based prompting to stabilize the simulation, betting that the developer community will uncover new use cases even as the underlying technology continues to evolve.

Comments (0)
No comments yet. Be the first!