Decart Launches Oasis 3: Photorealistic Driving Simulator Generating Hours of Content via API

Decart Releases Oasis 3: Photorealistic Driving Simulator Hours Through API

Decart, a startup in generative models of the real world, has opened API access to Oasis 3 — a system that generates photorealistic driving scenes in real time for testing autonomous vehicles.

What Oasis 3 Can Do

Oasis 3 is a so-called world model: a neural network that doesn't just draw static pictures, but simulates a dynamic physical environment. In practice, the model generates continuous video streams of driving through a city, highway, or parking lot — with other cars, pedestrians, traffic lights, and changing lighting conditions. The model works in real time and can synthesize hours of driving without stopping.

Developers get access through an API and can:

• Create custom routes and driving scenarios
• Test autopilot behavior in rare and dangerous situations
• Generate data for fine-tuning perception neural networks
• Check object detection systems without leaving the real road
• Scale testing without expanding the vehicle fleet

This can significantly reduce development cost and time: real-world road tests are expensive, slow, and legally complex in almost any jurisdiction.

Why World Models Matter for Self-Driving

Traditionally, companies developing autopilots choose between two extremes: real kilometers — expensive, slow, and requires a large fleet — or classical graphics simulators like CARLA. Simulators are fast, but their image is far from reality. Between these options lies a gap that engineers call the sim-to-real gap: a neural network trained on unrealistic data often gets lost in the real world, struggling with details that the simulator didn't reproduce.

Generative world models try to close this gap. Photorealistic images with reflections on wet asphalt, rain on the windshield, and sun glare give the autopilot data as close to real as possible.

"Our goal is to give autonomous vehicle developers a tool to scale tests without scaling the fleet," —

Decart noted.

Disclaimers and Limitations

Decart doesn't hide that Oasis 3 has weaknesses. Despite impressive realism, the system has known limitations:

• The model can produce visual artifacts during sharp lighting changes or extreme weather scenarios
• The accuracy of other traffic participants' behavior still lags behind specialized physics engines
• Generative scenes do not guarantee full authenticity of rare edge cases
• For safety-critical testing, data from Oasis 3 will require additional verification

Nevertheless, API availability means that small teams and startups get technology that was previously the domain of large labs like Waymo or Toyota Research Institute.

What It Means

Oasis 3 opens a new stage in democratizing autonomous system development: photorealistic synthetic data is now available through a simple API call. If world models mature to accuracy levels sufficient for production, they can fundamentally change the economics of automotive AI — part of expensive real-world testing will move to the cloud, and the barrier to entry for new players will drop significantly.