Decart launched Oasis 3 on June 10, a real-time world model that renders photorealistic driving environments on demand via public API, priced at $0.02 per second. The move puts the $4 billion Israeli AI startup in direct competition with the internal simulation efforts of every major autonomous vehicle company on the planet.
The pitch is straightforward: autonomous vehicle developers need millions of miles of edge-case training data they can never safely or economically gather in the real world. A child darting into traffic at 11 p.m. in heavy rain, a construction zone where lane markings contradict overhead signage, a truck jackknifing across three lanes. These scenarios are rare enough that real-world fleets might encounter them once in a billion miles of driving. Oasis 3 can generate them on demand, at scale, for fractions of a cent per second of simulation.
Decart framed the Oasis 3 launch as more than a product release. CEO Dean Leitersdorf, who co-founded the company in late 2023 alongside CPO Moshe Shalev after both spent time in Israel's elite Unit 8200, described it as the first world model people can actually program on top of. The framing is deliberate: Decart wants to be the infrastructure layer for physical AI the way OpenAI became the infrastructure layer for language. That is a large ambition, and the $300 million raise announced last month at a $4 billion valuation suggests investors are at least provisionally buying it. Radical Ventures led that round, joined by Nvidia, Sequoia, Benchmark, Adobe, and Toyota. Andrej Karpathy and members of the Nintendo founding family are among the angels.
The API pricing model is where the strategy gets interesting. By opening Oasis 3 to the public at a pay-per-second rate rather than locking it behind enterprise contracts, Decart is trying to replicate the developer ecosystem play that made OpenAI's API so sticky. The logic is that if enough startups, research teams, and AV companies integrate Oasis into their workflows, the switching cost becomes prohibitive before any of the incumbents can ship a comparable product. That is the theory, anyway. The practice is more complicated.
Waymo already has a world model. The company unveiled its own generative simulation architecture in February, built on top of Google DeepMind's Genie 3, and positioned it as a core training tool for the Waymo Driver. That system produces both camera and lidar output and lets engineers conjure scenarios through plain language prompts, a tornado on the highway, an elephant wandering into the road, situations no fleet will ever capture at scale. Tesla, Nvidia, and Wayve are building comparable systems in-house. Decart's real market is everyone below that top tier: the dozens of AV programs, robotics labs, and drone startups that want frontier-grade simulation without a DeepMind-sized research budget attached.
What two cents a second actually buys
Oasis 3 generates three synchronized camera feeds, one front-facing and two side-facing, which covers the perception setup of most camera-first AV stacks but stops short of the lidar output Waymo's system produces. As TechCrunch noted in its coverage of the launch, the model arrives with caveats, and the hard questions are the ones every world model faces: whether the physics hold up over long rollouts, whether generated edge cases actually transfer to real-world driving behavior, and whether safety regulators will ever accept synthetic miles as evidence of anything. The $0.02 rate is the public on-ramp. Enterprise pricing scales with the use case, and that is where the real revenue will sit if the strategy works. Decart says robotics and other physical AI applications come next, which widens the addressable market well beyond cars.
The next twelve months will show whether renting simulation becomes the default for physical AI or stays a niche for teams that cannot build their own. If mid-tier AV companies and robotics startups start shipping products trained substantially on Oasis-generated environments, Decart owns a layer of the stack nobody else is selling at retail. If the big players' internal models keep pulling ahead, two cents a second buys impressive demos and not much else. Watch the customer announcements, not the benchmarks.
Also read: A fired xAI engineer's Grok safety lawsuit arrives two days before SpaceX prices the largest IPO in history • Warner Music Group buys Sureel AI to own the technical stack that proves when artists' work trained an AI model • Cyera raises $600 million at $12 billion valuation as AI data security hardens into must-have enterprise infrastructure
