Xpeng's VLA 2.0 Drags Tesla's Crown Into the Real World
Xpeng testni automobil probija se kroz gusti pekinški promet dok senzorski tragovi prate vožnju.📷 AI-generated / Tech&Space, manual prompt only
- ★Xpeng drove 40 minutes in Beijing without a single human intervention, pushing the story beyond a simple demo clip.
- ★VLA 2.0 is an end-to-end vision-to-action model trained on 100 million clips from extreme driving scenarios.
- ★The proprietary Turing AI chip and Volkswagen as the first external customer turn the test into a signal for real industrial deployment.
Electrek reported that Xpeng's VLA 2.0 completed a 40-minute drive through Beijing traffic without a single human intervention. That is not a final proof of full autonomy, but it is a stronger signal than another carefully edited promo clip. Beijing is chaotic enough that one clean run in that environment matters. The key is not only the result on the road, but the architecture behind it. VLA 2.0 is an end-to-end vision-to-action model trained on 100 million clips from extreme driving scenarios, and Xpeng ties it to its proprietary Turing AI chip with 2,250 TOPS of compute. That matters because it is a cleaner industrial story than a stack of separate parts pretending to be one system. Here, perception and decision-making are built as a single loop. Another signal is Volkswagen, which the source says is the first external customer for the system. That means Xpeng is no longer talking only about an internal test, but about software that could move beyond one platform or one engineering team. In the automotive world, that can be a bigger step than the driving demo itself.
A 40-minute run through Beijing without human intervention sounds like a demo, but 100 million clips, a proprietary chip, and Volkswagen as the first external customer look more like the start of a serious program.
VLA 2.0 stack spaja kameru, računalni čip i odluke na cesti u istu vožnju.📷 AI-generated / Tech&Space, manual prompt only
The easiest mistake is to turn one clean test into a universal conclusion. One drive without intervention does not mean the system will behave the same way in other cities, weather conditions, or regulatory settings. Autonomy is not measured by one good lap, but by whether the same result keeps showing up as the environment changes. That is where this story gets interesting. Xpeng no longer looks like a company trying to catch Tesla's reference point. It looks like a company trying to connect its own chip, its own model, and its first external business signal into one commercial line. That is a more serious discussion than who has the better demo clip. The distinction between driver assistance and full autonomy still matters. VLA 2.0 can be a major step in advanced driving systems and still remain far from a universal answer for every road scenario. That is the point: Tesla is no longer the only name in the conversation, but Xpeng still has to prove that one good test can become a repeatable production standard. If that transition happens, 2026 could become the year the story shifts away from benchmark theater and toward the first repeatable industrial service built from model, chip, and road proof.
