Decart launches Oasis 3, a real-time world model for driving

On 10 June, Decart launched Oasis 3 — a real-time world model that generates photorealistic driving environments from natural-language prompts, with one front and two side camera views simultaneously. Type a scenario — fog, black ice, a flooded junction — and it spins up that scene, in any variation of weather and road surface, with no map to author by hand. Available via API at roughly $0.02 per second of simulated driving time. Primary market: autonomous-vehicle teams who need to rehearse rare and dangerous scenarios at scale. Next target: robotics and broader physical-AI applications.

CEO Dean Leitersdorf describes it as 'the first usable world model that people can actually program on top of' — pitching Oasis 3 as a platform layer for physical-AI development, not just a driving tool.

The catch. Multiple outlets that tested Oasis 3 flagged real-world limitations. For a driving training tool that is the central risk: a simulation gap can teach an AV the wrong behaviour, and that error shows up on a real road rather than in the test environment. 'Photorealistic' is not the same as 'physically accurate'.

The money. Oasis 3 follows a ~$300m raise at ~$4bn valuation — investors include Toyota, Adobe, eBay and Nvidia. Toyota's strategic involvement is particularly notable for the AV-simulation thesis.