src=”https://techcrunch.com/wp-content/uploads/2026/06/Oasis-a.png” />
AI startup Decart has introduced Oasis 3, its latest interactive world model engineered to generate photorealistic driving environments in real time. The model is currently accessible via API, a strategic move aimed at fostering a robust developer ecosystem akin to the approach taken with language models.
The Core Offering and Strategic Vision
Decart is initially targeting the autonomous vehicle sector, providing a scalable solution for simulating rare and complex driving scenarios. The company’s broader ambition extends to robotics and other physical AI applications. By offering API access from its inception, Decart seeks to empower developers to build directly on its foundational world model technology, anticipating the emergence of a vibrant community focused on this emerging AI paradigm.
“We believe this will be the first truly programmable world model,” stated Dean Leitersdorf, co-founder and CEO of Decart. “We anticipate a significant developer community will coalesce around this platform, driving innovation and expanding its utility.”
Building on the success of its real-time video model, Lucy, which has garnered a developer base of over 100,000 users primarily in e-commerce and livestreaming, Oasis 3 signifies Decart’s strategic pivot towards physical AI. The pricing model for Oasis 3 is set at $0.02 per second, with customized enterprise solutions available based on specific use cases.
Market Landscape and Competitive Dynamics
The release of Oasis 3 positions Decart within an increasingly competitive landscape of world model development. Notable advancements include Google’s research preview of Genie 3, World Labs’ Marble for commercial applications, and innovations from video generation startups such as Luma and Runway, who are also evolving their physics-aware video models into comprehensive world models.
This launch follows Decart’s recent achievement of securing $300 million in funding, a round driven by substantial demand for its existing models across various sectors, including physical AI. The significant capital infusion propelled the company’s valuation to nearly $4 billion, attracting strategic investments from prominent entities such as Toyota, Adobe, and eBay, all identified as potential future clients by Leitersdorf. Nvidia, an existing investor, also participated in this funding round.
Technological Edge and Efficiency Gains
Oasis 3 differentiates itself through its highly photorealistic output and infinite generation capabilities. This performance is attributed to Decart’s proprietary Decart Optimization Stack (DOS) software, which optimizes model execution across Nvidia, Amazon, and Google hardware. This vertical integration significantly reduces operational costs compared to competitors.
“Our deep integration with hardware, enabled by our real-time stack, allows us to operate these models at a fraction of the cost of industry peers—more than an order of magnitude cheaper,” Leitersdorf explained. He further highlighted the company’s capital efficiency, noting that Decart has utilized significantly less than $100 million in funding to date.
Performance and Current Limitations
Oasis 3 generates physically accurate, multi-camera environments (front and two side-facing) ideal for system training and testing. Unlike limited demos or research previews, Decart provides developers with unlimited scenario generation, a critical feature for autonomous vehicle developers needing to explore numerous edge cases. Initial assessments suggest Oasis 3 delivers superior photorealism and interaction depth compared to models like Google’s Genie 3 or World Labs’ Marble.
However, the model exhibits limitations in maintaining long-term environmental consistency. While initial scene generation is robust and adheres closely to prompts, thematic integrity tends to degrade during extended interactions. Testing revealed that environments can shift from specific locales (e.g., a New York City street) to generic urban settings, and previously generated intersections may disappear as the simulation progresses. Control responsiveness was also noted as an area for improvement, with occasional loss of directional control, a challenge shared by other leading world models.
A critical hurdle remains the simulation of physics, specifically regarding vehicle interactions. Instances where simulated vehicles pass through each other indicate an area where physics consistency requires further refinement. Leitersdorf acknowledged this as a significant research challenge, attributing it partly to the scarcity of data on accident scenarios compared to standard driving conditions.
The auto-regressive nature of Oasis 3, where each frame is generated sequentially based on previous outputs, contributes to this challenge and necessitates substantial computational resources. The model’s memory, currently limited by its context window, struggles to maintain long-term consistency as the number of generated frames increases.
“We are generating tens of frames per second, each represented by approximately 8,000 tokens, quickly filling the context window,” Leitersdorf elaborated. “Our research is focused on expanding this context to millions of tokens and developing more efficient memory compression techniques.”
Decart is exploring solutions, including enabling world generation from video input rather than static images, which may address some consistency issues in future iterations. Leitersdorf conceded that the field of world models is still in its nascent stages.
Developer Ecosystem and Future Outlook
Despite current limitations, Decart’s strategic focus remains on empowering its developer community. Leitersdorf draws parallels to the early days of large language models (LLMs) and the transformative impact of OpenAI’s API strategy.
“We envision a similar trajectory for Oasis 3, where developers will uncover novel use cases and applications that exceed our initial expectations,” he concluded. “I anticipate that within three months, we will witness hundreds of diverse applications built by developers leveraging Oasis, demonstrating its vast potential.”
Business Style Takeaway: Decart’s Oasis 3 represents a significant advancement in real-time, photorealistic world modeling, strategically positioning itself to capture the burgeoning physical AI market. The company’s focus on developer accessibility and operational efficiency through its DOS stack offers a compelling value proposition, particularly for industries like autonomous vehicles and robotics that require scalable, cost-effective simulation capabilities.
Based on materials from : techcrunch.com