Prime Highlight
- Nvidia launched Alpamayo, an open-source AI platform aimed at enabling human-like reasoning in autonomous vehicles.
- CEO Jensen Huang said the system helps vehicles handle rare, risky situations and explain their decisions clearly.
Key Facts
- Alpamayo 1 is a 10-billion-parameter vision-language-action model that uses chain-of-thought reasoning for safer driving decisions.
- Nvidia released over 1,700 hours of driving data along with AlpaSim, an open-source driving simulation platform.
Background
Nvidia on Monday launched Alpamayo, a new family of open source AI models, tools, and datasets that aim to bring human-like reasoning to autonomous vehicles. The announcement came at the Consumer Electronics Show 2026, where CEO Jensen Huang said the platform marks a turning point for physical AI.
Huang said Alpamayo allows machines to understand, reason, and act in the real world. According to him, the system helps vehicles think through rare and risky driving situations and clearly explain their decisions.
At the heart of the platform is Alpamayo 1, a 10-billion-parameter vision language action model that uses chain-of-thought reasoning. The model breaks a problem into steps, checks multiple options and then selects the safest path. This allows a car to deal with situations it has never seen before, such as navigating a busy junction when traffic lights stop working.
Nvidia said Alpamayo 1 does more than turn sensor data into steering and braking commands. It also explains why it chooses a specific action and predicts the vehicle’s future path, improving trust and safety.
The company has released Alpamayo 1’s code on Hugging Face. Developers can fine-tune it into smaller and faster versions, use it to train simpler driving systems, or build tools such as auto-labelling software that tags video data automatically.
Nvidia also introduced Cosmos, its generative world model system that creates virtual environments. Using Cosmos, developers can generate synthetic data and combine it with real footage to train and test Alpamayo-based systems.
As part of the rollout, Nvidia released an open dataset with more than 1,700 hours of driving data from different regions and road conditions. It also launched AlpaSim, an open source simulation platform on GitHub that recreates real driving conditions so teams can test self-driving software safely at scale.