Decart's Oasis 3 world model simulates photorealistic driving, but coherence fades

Decart, a two-year-old AI startup, has launched Oasis 3, an interactive world model that generates photorealistic driving environments in real time. The model is accessible via API at $0.02 per second, with enterprise pricing variable. Decart is targeting autonomous vehicle companies that need to simulate rare driving scenarios at scale, and plans to expand into robotics and other physical AI applications. The startup already has a developer community of over 100,000, built around its real-time video model Lucy. Oasis 3 is powered by Decart's DOS (Decart Optimization Stack) software, which optimizes models to run efficiently on Nvidia, Amazon, and Google hardware, making inference an order of magnitude cheaper than competitors. The company recently raised $300 million at a $4 billion valuation, with strategic investors including Toyota, Adobe, eBay, and Nvidia.

In practice, Oasis 3 delivers stunning initial scenes that match a text prompt — for example, a New York City street at morning looks photorealistic. But as users drive, thematic integrity degrades rapidly. The environment loses its specific character, and if you turn around, the original location is replaced by an entirely new scene. Controls also feel unresponsive, making the experience more like a disjointed dream than a coherent simulation. This limitation is common among world models tested by TechCrunch, but Oasis 3's infinite generation capability is a double-edged sword: it allows for extensive edge-case testing but amplifies drift. The model currently offers multi-camera outputs (front and two side views) and is positioned as the first 'usable' world model for developers, though its current coherence issues may temper excitement.