Artificial intelligence startup Decart on Wednesday unveiled the Oasis 3, its latest interactive world model that can create photorealistic driving environments in real-time, TechCrunch exclusively reported. The model is currently available via API.
THE startup it is initially aimed at autonomous vehicle companies that need to simulate rare driving scenarios at scale, and plans to expand into robotics and other natural AI applications. But the biggest bet is on developers: By offering API access from day one, Decart is trying to build a developer ecosystem on global models, just as OpenAI did with language models.
“It will be the first usable global model that people can actually program on top of,” Dean Leitersdorf, co-founder and CEO of Decart, told TechCrunch. “I think there’s going to be a whole community of developers that will emerge on top of that.”
The startup already has a community of more than 100,000 developers, many of whom are building products on top of the Lucy real-time video model, mostly in e-commerce and live streaming. Oasis 3 builds on this foundational model and represents the company’s push into natural artificial intelligence. Access is priced at $0.02 per second, and enterprise pricing depends on use cases, Decart said.
Decart plays in an increasingly crowded global model arena. Last year, Google released Genie 3 in research preview, Fei-Fei Li’s World Labs released Marble for commercial use cases, and video production startups like Luma and Runway are also translating their physics-aware video models into world models.
The launch of the Oasis 3 comes just weeks after two-year-old Decart raised $300 million, which Leitersdorf says followed “huge increases in demand for the models we built” in e-commerce, live streaming and physical AI. The round raised Decart’s valuation to nearly $4 billion and brought in a number of strategic investors including Toyota, Adobe and eBay. All of these companies are potential customers, says Leitersdorf. Nvidia, an existing investor, also participated in the round.
Oasis 3’s advantage lies in the photorealism of its models and infinite production capacity. This is due to some performance magic on Decart’s part, supported by the company’s other main product: its DOS (Decart Optimization Stack) software that allows models to run efficiently on Nvidia, Amazon and Google hardware, making its models much less expensive than competitors.
“This is built on top of our entire real-time stack, which we optimize down to the hardware,” Leitersdorf said. “By being so vertically integrated, we’re able to be more than an order of magnitude cheaper than anyone else in the industry to operate these models.”
The startup’s models are so efficient, according to Leitersdorf, that it has burned through “drastically less” than $100 million over its lifetime.
Oasis 3 creates physically accurate environments with multiple cameras — one in front and two on the side — for training and testing systems. And instead of offering limited demos and research previews, Decart allows developers to create scenarios without limits, which is perfect for autonomous vehicle developers who want to test as many extreme cases as possible.
Compared to other models I’ve tested, such as Google’s Genie 3 or World Labs’ Marble, the Oasis 3 offers the most photorealistic environments from a single text message I’ve seen. And the fact that you can interact with them for hours suggests a level of efficiency that Decart’s rivals may lack.
But letting you create a world for so long also degrades the model significantly.
In my tests, I found that the system could consistently create a strong opening scene that matched the prompt, but the thematic integrity quickly degraded as I moved the world around. I pushed it to generate a New York street in the morning, it did, beautifully. But as I drove, the environment felt less like New York and more like a standard version of any urban, western city.
When I tried to turn around and go back to the original intersection, it was gone, replaced by a completely new environment. Additionally, the controls aren’t very responsive and I often lost control of where the car was going (again, a shortcoming shared by other global models I’ve tested). The experience felt less like a coherent simulation and more like a dream, a disjointed stream of consciousness that quickly turns nonsensical.
Another issue, which I’ve seen in other world models, is that the car will pass other cars, meaning the model doesn’t properly simulate the physics in the environment. Leitersdorf calls this a “major research problem we’re solving now,” attributing it to the fact that “there is drastically more data on good driving compared to accidents.”
Part of what makes this consistency of physics difficult is fundamental to how this global model works. Oasis 3 is automatically regressive, meaning it creates one frame at a time and looks back at what it has previously created to decide what to do next. This is a key architectural feature of many global models, and it is also computationally intensive.
In order to maintain consistency, Leitersdorf says Decart’s team is working to improve the model’s memory length.
“Each frame we make is about 8,000 brands,” he said. “Rendering this at tens of frames per second — that’s hundreds of thousands of tokens per second. The context window fills up very quickly. We’re researching how to make a larger context to store millions more tokens and how to squeeze memory into fewer tokens.”
Leitersdorf believes the consistency issue may be partially resolved in the next version of the model, which will allow users to start creating worlds based on a video of an environment rather than an image. He acknowledged that global models as a field is still early days.
However, the founder is less focused on the current limitations of his technology than what will happen when developers get their hands on it.
“It takes me back to the early days of LLMs, when OpenAI invented the API for models,” he said, pointing to the emergence of a community of developers that advanced the field by finding and creating new use cases.
“When we talk again in three months, we’ll say, ‘Here are 100 developers who all built 100 different apps with Oasis that blew us all away,'” he said.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
