In the bustling heart of Beijing, where traffic is an unpredictable dance of aggressive drivers and chaotic lanes, the XPeng VLA 2.0 emerges as a game changer in the realm of autonomous driving technology. Over the past few years, the electric vehicle (EV) market has seen a surge in competition, challenging Tesla's longstanding dominance. XPeng's latest offering, with its innovative Vision-Language-Action technology, aims to redefine what we expect from self-driving cars in today's challenging urban environments. This article delves into the performance of the XPeng VLA 2.0 amidst the frenetic pace of Beijing’s streets and its aspirations to rival established competitors.
Beijing traffic does not care about your feelings. It is a place where lane lines are merely decorative. Drivers there move with a level of aggression that would make a New York City cab driver blush. If a car hesitates for a fraction of a second, three other vehicles will fill that space. This is the ultimate testing ground for electric cars that claim they can drive themselves.
I went to Beijing last week to test the new XPeng VLA 2.0 and witness it take to these chaotic streets. The results suggest that the era of Tesla being the only serious player in the self-driving game is officially over.
XPeng started sending out its VLA 2.0 software through over-the-air updates in March 2026. It went straight to the company's "Ultra" versions of the XPeng P7, XPeng G7, and XPeng X9. VLA stands for Vision-Language-Action. The 2.0 version represents a monumental change in how EVs perceive the world.
Older systems operated like a group of people who hardly communicated with each other. One segment observed the road, another planned the route, and a third manipulated the wheels. This often led to delays or jerky driving. VLA 2.0 utilizes a single "brain" that processes camera images and decides on actions immediately. This streamlined system makes driving feel significantly more fluid.
To facilitate this, XPeng developed a powerful dedicated computer chip, the Turing AI, capable of handling 2,250 TOPS of computational power. The system underwent extensive training, utilizing 100 million video clips of highly challenging driving scenarios. This training enables the VLA to function more like an experienced human driver rather than a stiff robot.
During an initial 40-minute test drive in the heart of Beijing, the system performed remarkably well without requiring any human intervention. In a city characterized by cutthroat competition among drivers every thirty seconds, the car demonstrated composure, facilitating smooth navigation through busy intersections and overcoming aggressive road users effortlessly.
This accomplishment was not a fluke; two additional test drives followed, accumulating over four hours of driving. One of these drives also traversed the bustling streets of Beijing, while the other embarked on a spontaneous journey through rural roads devoid of clearly marked lanes and featuring even more unruly drivers. Throughout both testing phases, there was no need for manual interventions — a rare feat in modern EVs.
One particular moment stood out during the drives: merging into a tight gap amidst fast-moving traffic. Most driver-assistance systems from other EV brands would either hesitate for a substantial opening that may never materialize or relinquish control to the human driver. The XPeng P7 Ultra, however, calculated the speed of surrounding vehicles and confidently darted into the narrow gap, executed smoothly and drawing parallels with the assertiveness of a seasoned local driver.
In another test, the car successfully navigated a challenging merge from a rural road to a busy five-lane thoroughfare bustling with traffic coming from multiple directions. The vehicle sensed the chaotic environment and navigated the situation adeptly, demonstrating finesse as it positioned itself before smoothly joining the fray.
The P7 even exhibited a cautious and prudent behavior when approaching a blind turn. It slowed down effectively, creating enough space for a scooter dodging bumps and potholes. Emulating the actions of an experienced driver, the car held back, scanned the surroundings, and once it ensured a safe gap, it advanced confidently, prioritizing the safety of all road users ahead.
This driving experience was almost surreal, reminiscent of watching a seasoned driver in action—exhibiting no hesitation, possessing total awareness of the surroundings, and leaving ample space in potential scenarios requiring sudden reactions.
Another interesting aspect was the car's behavior on narrow, unmarked roads, where it maintained a centric position instead of rigidly hugging the right edge. This allowed it to deftly navigate around bumps while remaining responsive and prepared for any oncoming traffic.
Such human-like sophistication results from a concept dubbed the "generative world model." This virtual reality model allows XPeng to conduct tests of its software without actual road trials. The company has ramped up its scenarios from 30,000 annually to 500,000 per year. Each day, the system simulates the equivalent of 30 million kilometers, allowing the car to learn from mistakes within a digital environment before hitting the actual streets—contrasting with Tesla's reliance on real-world data.
The rivalry among EV companies is starting to take on dramatic overtones. XPeng's CEO, He Xiaopeng, recently visited Silicon Valley to test Tesla's FSD v14.2, spending five hours driving in San Francisco and offering praise for Tesla's system.
He has set an ambitious target for his team, striving for the VLA system to match Tesla’s best performance by August 30, 2026, even humorously betting with his head of autonomous driving on the outcome.
Tesla faces challenges in the Chinese market as well. While North American users enjoy the latest software, Tesla drivers in China remain with an older version due to slow government approvals. This situation presents an advantageous edge for local brands like XPeng.
Also, competitors like BYD are already integrating innovative driving systems into EVs priced around €26,000. Tech titan Huawei is investing over €9 billion into developing its self-driving technology, and Xiaomi is also entering the fray with its SU7 lineup, further saturating the competitive landscape.
On the business front, the landscape is shifting as Tesla charges a monthly subscription fee of €85 for its full self-driving capabilities. In contrast, XPeng integrates its VLA 2.0 system at no extra cost within the vehicle price, making it challenging for Tesla to justify its subscription model when competitors offer similar capabilities for free. Volkswagen has recognized XPeng's prowess, signing a deal to implement VLA 2.0 in its electric SUV designed for the Chinese market—a significant endorsement from an established automotive name.
One standout aspect of the VLA 2.0 is its predictive capabilities over mere responsive behavior. Unlike many existing systems that react to immediate obstacles, this system anticipates traffic patterns, positioning itself for turns and merges ahead of time. This forward-thinking technology alleviates the hesitancy common in various driver-assist systems.
Many competitors seem to hesitate or guess their next move, leading to passenger unease. In contrast, the VLA 2.0 operates with a clear sense of planning. When it does make a minor error, such as following too closely or hesitating at a junction, it quickly adjusts itself in a safe manner.
We are undoubtedly at a crucial juncture in the automotive world. Consumers once turned to Tesla for its exclusive "smart" features, but now, the XPeng P7, XPeng G7, and XPeng X9 have demonstrated comparable technological capabilities. The forthcoming XPeng GX is set to raise the bar even further with the potential for Level 4 autonomy. The question of whether XPeng can officially claim superiority over Tesla necessitates further head-to-head evaluations.
For those traversing the challenging terrain of Beijing traffic, the VLA 2.0 system stands out as a robust companion. If it can navigate Beijing's complexity safely, it will likely excel in traffic conditions across the globe (with perhaps a few exceptions, such as India). The gap in autonomous driving technologies is closing rapidly, leading to an exhilarating chase in the future of electric vehicles.
