Nvidia has released Alpamayo 2 Super, what it calls its most powerful open AI model yet. The model is designed specifically for robotaxis — self-driving taxis that operate without a human behind the wheel.
What Alpamayo 2 Super does
The model handles perception, prediction, and planning tasks that autonomous vehicles need to navigate roads. It can process data from cameras, lidar, and radar in real time. Nvidia says the model is open, meaning developers can download, modify, and integrate it into their own systems without paying licensing fees.
That openness is a shift. Most companies developing autonomous driving technology keep their models proprietary. By releasing Alpamayo 2 Super under an open license, Nvidia is betting that the broader developer community can help improve and deploy the technology faster than any single company could.
Why robotaxis are the target
Robotaxis are one of the hardest challenges in AI. The vehicle has to handle unpredictable traffic, pedestrians, weather, and road conditions. It has to make split-second decisions that affect safety. A model that works in a controlled test track often fails in the real world.
Nvidia has been supplying hardware and software for autonomous vehicles for years. Its Drive platform powers systems in cars from companies like Mercedes-Benz and Volvo. Alpamayo 2 Super is the first model from Nvidia built from the ground up for the robotaxi use case, rather than adapted from a general-purpose system.
The model is named after Alpamayo, a peak in the Peruvian Andes. Nvidia has a habit of naming its AI models after mountains — the previous generation was called Alpamayo 1.
What openness means for the field
Open AI models have become more common in areas like text generation and image recognition, but they are still rare in autonomous driving. Safety concerns are one reason. A flawed model could cause accidents. Liability is another. If an open model causes a crash, who is responsible?
Nvidia is not releasing the model as a finished product. It is providing the underlying neural network architectures and training code. Developers will need to adapt the model to their own vehicles, sensors, and regulatory requirements. The company has also published a technical paper that describes how the model was trained and tested.
The release includes benchmark results showing Alpamayo 2 Super outperforming its predecessor on standard driving scenarios. Nvidia did not provide independent verification of those results.
The model is available now through Nvidia's developer portal. Robotaxi startups and established automakers can start testing it immediately. The first vehicles using Alpamayo 2 Super are not expected to hit public roads until later this year.
