At the 2025 XPeng Tech Day, XPeng Motors introduced its second-generation Visual-Language-Action (VLA) model. This major release not only broke traditional industry model architectures but also achieved end-to-end output from visual signals directly to action commands, freeing itself from the constraints of language translation. XPeng's innovative approach has made it the first physical world large model to be mass-produced, providing a strong intelligent foundation for multiple fields such as autonomous driving, artificial intelligence robotics, and flying cars.

The capabilities of the second-generation VLA are impressive. It is equipped with billions of parameters in the Ultra version vehicles with 2250TOPS, while the parameter count of similar industry vehicles is generally only in the tens of millions. This means that XPeng's technology far exceeds competitors in terms of computing power and model complexity. To achieve this breakthrough, XPeng collected and utilized nearly 100 million clips of data, equivalent to the total number of extreme scenarios experienced by a human driver over 65,000 years.

Artificial Intelligence Driving

This new model is not just an action generation model, but also an intelligent system that understands and infers the physical world. Its application range is extensive, capable of driving various types of artificial intelligence devices across domains. Thanks to the second-generation VLA, XPeng also launched "Xiao Lu NGP," which has improved the average接管 mileage on complex small roads by 13 times. At the same time, the first "No Navigation Automatic Auxiliary Driving" Super LCC + is highly anticipated, allowing global users to drive safely without navigation.

He Xiaopeng, founder of XPeng Motors, stated that in the future, large models will become the "operating system of the physical AI world" and will promote the development of all physical AI applications. The release of the second-generation VLA marks another important progress for XPeng in the field of intelligent driving and takes a solid step toward embodied intelligence.

The second-generation VLA will begin pilot co-creation experiences in December 2025 and is expected to be fully rolled out in the Ultra model in the first quarter of 2026. Meanwhile, Volkswagen has become its first strategic partner, laying a good foundation for their future cooperation.