I believe 2026 will be a year of great success for everyone, right? We just saw a little horse, right? I also believe it will be a magical and lucky year. So today, our opening is a very interesting one; we've invited our magician, Mr. Guo, to help us open the show. Okay, back to today's press conference. Welcome everyone, welcome to all our online and offline friends, welcome to all our friends from China and around the world, welcome to the 2026 XPeng Global New Product Launch Conference! I believe this is a brand new beginning. So this year, XPeng is bringing you a very new press conference at the beginning of January. Actually, as everyone knows, January rarely hosts new product launches. But I believe that 2026 to 2030 will be a magical five-year period. I believe that in these five years, all global car manufacturers, robot manufacturers, and physical technology manufacturers will experience a new year and a new look. This is also a magical new beginning for XPeng. Before we begin 2026, I'd like to quickly share and report on XPeng's plans for 2025. First... XPeng delivered a total of 429,000 vehicles globally in 2025, a year-on-year increase of 126%. We achieved a very good result. We also reached the milestone of over 1 million cumulative deliveries in 2025. I believe the next million is not far away. Meanwhile, XPeng entered 60 countries and regions globally in 2025. From January to December, we delivered a total of 45,008 vehicles in overseas markets, a year-on-year increase of 96%. Moreover, we ranked first in sales among new energy vehicle brands in 10 countries. Let's look at the sales figures for all new energy vehicle brands globally. Because today's sales figures only cover October, and due to differences in insurance registration and other data lags in different countries, we see that from January to October 2025... I believe what I'm most concerned about isn't XPeng's ranking, but rather how many Chinese brands are represented. We've counted them all—there are 7 Chinese brands, right? So I think we should all applaud the global expansion of 'Made in China.' I believe that in the next decade, everyone will see significant progress, whether in the internet, mobile internet, AI, or automotive sectors. In the field of low-altitude robot physics and AI, we are seeing more and more Chinese companies going global and joining the top tier. I also believe that XPeng's eighth-place ranking will only continue to rise. In the past few years, we have seen XPeng making continuous efforts globally, but we will accelerate in 2025. By 2025, we will have nine global R&D centers, and we will start global localized production for the first time. So last year, we saw that XPeng had three overseas production bases outside of China, including Austria in Europe and Indonesia and Malaysia in Asia. We believe that in 2026 and beyond, we will have more global production bases. We hope that everyone around the world can see and drive the best XPeng cars. In the past few years in China, we have been building a charging network. I am very happy that, as of now, we have connected to 3,150 self-operated supercharging stations and 2,650 super-fast charging stations in China. At the same time, we have connected to a large number of third-party cooperative charging stations both in China and overseas. I would also like to announce here that in 2026, XPeng will officially not only deploy its own ultra-fast charging system in China, but we will also begin deploying XPeng's self-operated supercharging stations in Europe, Asia, the Americas, and many other global regions. In 2025, we also launched the Kunpeng Super Extended Range globally. This signifies XPeng's official entry into the era of dual-energy vehicles. We started this era last November. In fact, we hesitated for a long time about creating a vehicle with both pure electric and super extended range capabilities. In the past few years, when we went global, we saw that many countries' charging infrastructure was far behind China's, and their development speed was far behind. Many countries lacked good electricity, good roads, and even stable networks. Therefore, XPeng's vehicles, whether in terms of dual-energy or local AI, have considered global development. So, I want to share with you that in the past decade, XPeng was a representative of intelligent electric vehicles. In the new decade, XPeng is a representative of physical AI plus globalization. In 2025, XPeng entered the initial deployment of physical AI. Starting in 2026, I believe everyone will see XPeng become one of the first Chinese and even global automakers to fully embrace and begin... For companies entering the era of emerging physical AI, XPeng will see the implementation and mass production of physical AI in 2026. From research and development to product development several years ago, mass production will truly begin in 2026. In November 2025, at our Technology Day, we announced four physical AI-related initiatives, all of which will be mass-produced in 2026. We expect to see the second-generation VOA and our robotaxi implemented and mass-produced in the first half of this year, and our humanoid robots and flying cars will be implemented and mass-produced in the second half of the year. Many of our offline friends may have visited our flying car showroom today. Flying cars have been a part of XPeng's business for 13 years. Imagine, it took 13 years to develop an idea, creating seven different prototypes before finally reaching mass production. This is a huge challenge for us. However, I believe that more and more companies and more amazing technological products will truly enter our lives. Furthermore, based on globalization and our underlying physical AI capabilities, I am very happy to see the changes in four models presented to us today. While preparing for today's press conference, we noticed that CCTV had just released the new theme for the Year of the Horse Spring Festival Gala, called 'Miracle on the Road.' We felt that every word of that theme resonated with the meaning of each car we're launching today. I think this symbolism is particularly auspicious. So, I wanted to allow everyone to experience the characteristics of XPeng's four cars while also feeling the festive spirit of 2026. Which four cars? First, there's the XPeng P7 Plus, a global model with dual capabilities. Then there's the globally-oriented PG Plus, which I'll share in detail about its upgrades and improvements later. Also today, we'll be introducing the XPeng G7, launching a super-range extended version. The G7 is XPeng's first model equipped with its self-developed Turing AI chip. Today, we not only have the pure electric G7, but we're also adding a super-range extended version, allowing more regions in China, including the western, northern, and central-western regions, to experience XPeng's unique performance in intelligent family SUVs. Finally, there's the XPeng G6 pure electric, a six-in-one all-electric coupe SUV. They will also have a refresh today, including XPeng's G9 pure electric AI intelligent luxury SUV. This is a vehicle with excellent range technology capabilities. Today, we will see its refresh. This 'miracle driving' refers to the four models that XPeng will be releasing today. I will introduce them to you one by one. Before introducing these four models, I would like to first introduce XPeng's second-generation VOA, which will be equipped in all four models. It is also the industry's first physical world model to achieve L4 level initial capability. In fact, in this field, we have been working on end-to-end VUA since 2023, and we started developing VUA in 2024. By the beginning of 2025, we saw the second-generation VUA, which can truly achieve L4. The first-generation VA was at most L2. We ultimately chose to skip the first generation and go directly to the second generation. I am particularly happy that we have achieved an idea that seemed impossible to realize at the end of 2024 or even the beginning of 2025, and we have achieved it this year with our recent extremely high growth rate. A few weeks ago, I wanted to share a story with you. I was in Guangzhou at the time, continuously driving different test versions of our VA2.0. I also went to Silicon Valley and experienced Tesla's latest FSD V14.2. To be honest, I can only describe it with one word: shocking. Everyone knows there are all sorts of sensationalist headlines saying things like 'Shocking!', right? But I really want to say that before, I always thought that both NGP and FSD were just Level 2, just a slightly better version of Level 2. That was my initial impression. So, in the past, when I experienced different city driving assistance systems from Chinese and overseas companies, my feeling was that they were all pretty much the same in the end. They were usable, but you had to be careful using them. They all had various advantages, but also various disadvantages, and you couldn't be completely confident in them. But honestly, after experiencing FSD in Silicon Valley, I was extremely shocked. I saw Xpeng skipping from VVA1.0 to VVA2.0. I think our speed of change is very fast. I think we were among the first to embrace a system that satisfies both Level 4 and Robotaxi users. But I was also really surprised to see Tesla's FSD in this upgrade, because in V13 it was still completely Level 2. However, V14 uses the same hardware and computing power, but it's upgraded to a completely different level. I believe that most of you here today, both online and offline, haven't truly experienced it. Because if you haven't experienced it, you won't have that feeling of a completely new era beginning. You won't feel the difference between the previous generation and the next generation of cars. You really can't imagine it just by reading words without experiencing it firsthand. So, from another perspective, you can see a few short videos here. From a certain perspective, I think Tesla's FD and VA are improving by 5, 10, or even dozens of times. And recently, we've seen something interesting: Waymo's L4 self-driving cars were paralyzed during the recent power outage in San Francisco. You might not know that L4 cars actually have a 'God's Eye' and a 'God's Hand.' What is a 'God's Eye'? Because it has a high-definition map. You can see that many L4 cars, when going anywhere in the city, first need to scan the map, and it won't dare to drive in places it hasn't scanned. That's why L4 cars can't go to residential areas, industrial parks, or underground parking lots—because they haven't scanned the map. All their routes are not based on real-time observation. It combines the 'eye of God' with an invisible 'hand of God.' This means that many L4 vehicles claim very little or no overtaking over even after 100,000 kilometers. If you drive one and observe, for example, a car parked on the road suddenly pulls over. It doesn't move for a few seconds, then suddenly changes lanes and moves. Why? Because the 'hand of God' takes over from the cloud. You just don't know that we might use many safety drivers remotely to take over. It's possible in large cities, but not in medium or small cities. Therefore, the generalization ability of L4 vehicles is often poor because there are too many scenarios in the world. After XPeng developed robots, because the scenarios for L4 autonomous driving are, in my opinion, 1,000 times more complex, you'll find that there's no way to write rules using algorithms. That era is over; it's the previous era. Just like when we first encountered the internet in the 1990s, you'll find that digital information was completely transformed by the internet. So, what I especially want to say is that when you actually have L4, don't rely on imagination; you must actually use and experience it. Only true full L4... A truly valuable point is one that can be implemented in all scenarios. So, going back a year ago—and today is early 2026, but I was in early 2025—I was in the second year of our end-to-end autonomous driving R&D. In the third year, after two years of development, I suddenly realized that one day, while driving our car, our product director showed me a new version, and I found that this end-to-end model did indeed have better generalization than previous autonomous driving systems. However, I felt it could never truly achieve autonomous driving because it was like a seesaw: its efficiency improved, but the user experience declined; its safety improved, but its capabilities declined. Secondly, I think it's a huge Frankenstein's monster. Many people talk about end-to-end systems—one-stage, two-stage, multi-stage—but they are all at most data-driven end-to-end, not truly model-driven end-to-end. So, the problem I see is that you'll encounter many issues where things work today but not tomorrow. Why? Because it contains a large number of rules to cover potential problems. These issues mean it can never truly function in new scenarios. Therefore, we must tear down the old structure, rebuild the foundation, and create a completely new autonomous driving system. So, what makes me particularly happy is... XPeng is working on a second-generation VLA to find better solutions to the infinite complexity of the physical world. Therefore, we've developed a different kind of large model, with different computing power and different data. Many people say the model is huge. For example, today, most companies using end-to-end autonomous driving probably have models smaller than 0.5 bytes – a very small model. More importantly, most of the models they actually use are still language-driven models focused on the digital world, rather than models based on the physical world, its appearance, and motion. Therefore, their model conversion efficiency is lower. Only by using entirely new models with larger parameters can this be possible. Secondly, it requires enormous computing power. Many people say, 'I can just buy huge computing power.' But only after XPeng developed, mass-produced, and implemented our new computing power last year did we realize that this 'large computing power' has a different interpretation from our previous understanding. It's not just about high computing power; you need the ability to use that computing power effectively. I'll say something that might not be entirely accurate: I believe you can't fully utilize computing power by relying on third-party chips. Only with your own chips, your own compiler, your own quantization, your own English and French syntax, and your own models adapted to the computing power can you achieve this. Only then can the same computing power be made to exert one to several times the power of others. So today, I believe that XPeng's computing power far exceeds that of its competitors. If it's 2250 TOPS, we may have reached 4500 TOPS. The third is the larger amount of data. Many people used to say that data is valuable because we can have data on every car, such as what kind of car is in what scenario. So XPeng now uses 100 million different data points. Remember, data from driving on the same road is worthless because it has been repeated countless times. Only 100 million CLIPs of data under various abnormal conditions are equivalent to a human driver learning from 6,500 different situations. Think about it, would you drive for 6,500 years in different scenarios? You wouldn't. Only by combining the data from these scenarios can we achieve L4. So from our perspective, there are four very core differences between L2 and L4. The first is capability. In the past, for example, many companies said that they had NOA in the city. There are many types of roads in the city, including main roads, side roads, and industrial parks. Many only drive on a few expressways. This is not considered a significant achievement. This XPeng NGP can drive on city expressways without much difficulty. But the most important thing is your ability to handle various situations. For example, can you make an unprotected left turn? Many people can do that. But if someone is crossing the road and you're waiting for a red light, and the light is about to turn green, should you proceed? You should pay attention to the road conditions and their right of way, and stop. So, how do you control your car in this situation? There are many different abilities involved in different scenarios. For example, can you handle different scenarios, from main roads to expressways to side streets to parking lots to residential areas? In the past year, if you've experienced all the assisted driving systems in China, such as those in parking lots, side streets, and residential areas, you'll know that they are basically memory-based driving systems. That is, they drive very slowly and rely entirely on memory logic. They can't actually handle all scenarios smoothly, safely, and quickly, and can't handle all kinds of cases. Third, what is Anxin? It's an upgraded version of safety. For example, if you drive a car for 100 kilometers without needing to take over, that's safe. But Anxin is different. When making a fast turn, you won't feel any collision with the car next to you. When there are many people or small motorcycles passing by, you feel at ease, thinking you won't collide with them. When there's a mud truck next to you, you feel your car is a little too far away from the mud truck, and you don't need to worry. What does it mean to feel at ease? It means that even parents unfamiliar with autonomous driving can sit in it without feeling scared. That's true peace of mind. Fourthly, efficiency is key. Recently, I've experienced some end-to-end assisted driving systems. I found that they can drive, but very slowly. They stop everywhere and then move slowly, or even not at all. This isn't how humans drive. We humans must balance capability, scenario, safety, and efficiency. So, I want to tell everyone that my goal for VA2.0 is that when it's released, the end-to-end autonomous driving take-off capability will be at least 5 times higher. That is, the take-off rate should be 1/5 of the current level. I hope our scenarios will not only allow driving on main roads and expressways, but also on small roads, in residential areas, and parking lots, enabling stationary starts and park roaming. I hope my sense of security will allow me to truly feel safe even with my hands on the controls, in any corner case, even if it complies with traffic regulations. Regardless of day or night, regardless of any anomalies, I'm ultimately a very efficiency-driven person. Many people who eat with me suffer because I finish my meal in five minutes, then sit there watching my friends eat. It's about being 5% faster than driving; I think that's more efficient. So today, I'm sharing some short videos of myself driving the VRA beta version.
I've been driving VOA a lot lately because every time they release a new version or I go to a new area, I use full navigation. It's very reassuring. Then I observe how others drive, because their methods differ from mine, leading to many interesting points. From another perspective, what we just saw is incredibly smooth handling of irregularly shaped vehicles and various construction stilts, without any delay, lane changes, driving against traffic, and allowing oncoming cars to return to the main road. This also includes defensively slowing down in the S-curve of a blind spot, as seen earlier. We can see that other related driver assistance systems either have to take over or crawl at very slow speeds. When merging from a side road onto a main road, it achieves both safety and right-of-way control, with an invisible process that perfectly balances efficiency, peace of mind, and safety. There's something very interesting here: you see, while we're driving, there's someone next to us—I don't think this person is a traffic policeman, but just a construction worker in a yellow uniform—who, when a construction vehicle passes by, tells you to stop and let the vehicle pass by, and waves. The vehicle understands. We absolutely didn't write any rules down, so at this point, it really needs to learn from a large amount of data; it understands the rules on its own. Now, I want to share some good news with you. We took this assisted driving system to Europe without any training—remember, without any data training—and it performed exceptionally well. Although there were a few places where the traffic regulations weren't understood, the performance on the main roads and side roads we saw completely exceeded our expectations. Therefore, we are very confident that this automated assisted driving system will not only be effective in any scenario in China, but we believe that globally, we will also conduct further post-training and enhancement using local data, in accordance with laws and regulations, to launch it. I believe you will see XPeng in China in 2026. In 2020, XPeng's autonomous driving technology saw significant improvements globally. This is something we're currently researching and developing; I've already seen their test vehicles, and the results are truly excellent. Imagine starting from a standstill and telling it, 'Go home,' it drives itself out of its parking space. It doesn't slowly and sluggishly drive out; it smoothly pulls out and pulls over. It recognizes various speed bumps and parking spaces in China—I think Chinese parking lots are even more challenging than American ones, to be honest. There are also different architectural styles. For example, if you're in a hurry and want to drive faster, or if you don't know your car well and prefer standard mode, or if your mother wants to drive slower, you can achieve different driving styles. So, from another perspective, I believe that when XPeng's autonomous driving technology is released, every version upgrade will be amazing. So, last year, when I was very confident, I made a bet with our large autonomous driving team in Silicon Valley, Guangzhou, and Shanghai: by August 30, 2026, could we achieve autonomous driving capabilities similar to Tesla's in the US? In fact, if we achieve this, because as everyone knows, China's area is much larger than Silicon Valley, and China's roads are much more complex than Silicon Valley's, I believe this capability will definitely be improved several times over. But I still hope we can achieve it. So at that time, I made a bet with our head of autonomous driving, Xianming. I asked him if he dared to bet. Xianming was very confident at the time. He said he dared. But today, Xiemin is here from the United States on a business trip. I want to call out to Xiemin, Xiemin, how confident are you that you won't run it naked?
