Technology
Decart's world model is real — and narrow on purpose
It dropped the sci-fi 'simulate anything' pitch for something that might actually sell.
Decart's Oasis 3 (10 June 2026) is a real-time driving world model via API — focused, not universal.
Launched 10 June: a world model that generates photorealistic, promptable driving scenes in real time, multi-camera, via API at ~$0.02/sec. Target buyer: AV teams who need to rehearse rare, dangerous scenarios at scale. CEO Dean Leitersdorf wants developers to build on it like the early OpenAI ecosystem — platform ambition under a focused product.
The caveat that actually counts
Outlets that tried it flagged real-world flaws. Normally 'demo has rough edges' is forgivable. Here it's the whole ballgame: if you're training a self-driving car against a simulation, the gaps between sim and reality are precisely where cars crash. A world model that's 95% photorealistic can teach a 5% wrong lesson, and you find out on a real road.
So: genuinely interesting, appropriately scoped, and don't let 'photorealistic' do too much work. The bet that pays off isn't the pretty driving clip — it's whether developers build enough on top that Decart becomes the layer everyone else's physical-AI uses. Long road. Real start.
Sources
- Decart's new world model can simulate hours of photorealistic driving — with some caveats — TechCrunch, 10 June 2026
- Decart Lays The Foundation For Physical AI Systems With Oasis 3 — Dataconomy, 10 June 2026
- Decart launches Oasis 3 world model for robotics and autonomous vehicle training — Robotics & Automation News, 11 June 2026