At the 2026 Xpeng Global New Product Launch, a breakthrough in the fundamental logic of autonomous driving was officially unveiled. Xpeng Automobile Chairman He Xiaopeng announced that the Ultra versions of the 2026 P7+, G6, G7, and G9 will be fully equipped with Xpeng's self-developed second-generation VLA (Vision-Language-Action) large model—this is not only the first industry model to claim L4-level capabilities, but also marks a shift in the intelligent driving system from the "perception-decision" paradigm to a new "understanding-inference-generation" paradigm.

Different from traditional autonomous driving systems that rely on rules or limited scenario training, the second-generation VLA is essentially a unified intelligent agent capable of understanding, inferring, and generating actions. It does not simply react to the current environment, but can actively simulate traffic dynamics for several seconds or even longer periods based on massive real-world data, anticipate potential conflicts, and generate optimal behavioral strategies. Particularly importantly, the model can be trained end-to-end using nearly 100 million real driving video clips without manual annotation, achieving truly self-evolving learning.

This capability gives it a significant advantage in handling long-tail scenarios—those rare yet critical edge cases. By internally generating adversarial long-tail scenarios and repeatedly training, the VLA model can continuously "rehearse" extreme situations, thus more calmly resolving crises in reality. Furthermore, the model has cross-domain driving capabilities, meaning its intelligent core is not only suitable for sedans but can also be seamlessly migrated to SUVs, sports cars, and even future flying cars, building a unified intelligent mobility ecosystem.

Alongside the VLA, two key feature upgrades were also unveiled. The first is "Xiaolu NGP," specifically designed for complex urban side roads, narrow alleys, and roadways without markings—areas traditionally beyond the reach of intelligent driving systems—greatly expanding the coverage of high-level assisted driving. The second is the upcoming "No Navigation Automatic Driving"—the Super LCC+ human-machine co-driving system. This means the vehicle no longer relies on high-precision maps or pre-set navigation routes, but instead can achieve point-to-point intelligent travel on open roads by relying solely on real-time perception and the VLA's inference capabilities, truly moving towards the ultimate goal of "drivable anywhere across the country."

The newly launched models each have clear positioning: the P7+ continues to maintain its status as an intelligent sports sedan flagship, the G6 targets young tech-savvy families, the G7, as a new mid-to-large SUV, fills the gap in the premium market, while the G9 further strengthens the position of luxury intelligent flagship. However, their common soul is this second-generation VLA model. Xpeng is using the "physical world model" as a lever to shift the technical focus of the entire intelligent driving industry—from competing in sensors and computing power—to competing in the ability to understand and generate real-world scenarios. When cars begin to think, simulate, and act like humans, L4-level autonomous driving may no longer be far away.