MG 07, a young Tesla in China?

During the 2026 Beijing Auto Show, the MG brand announced that its MG 07 (parameter | inquiry), a heavyweight model of electro-intelligent transformation, will be equipped with the first batch of R7 reinforcement learning world model + XHEART X7 dedicated large model chip, which is a "pure-blood" Momenta R7 solution. After the auto show, SAIC MG brand division general manager Chen Cui, Momenta CEO Cao Xudong and other core media exchanged. Chen Ji revealed in the exchange that the first batch of vehicles in this "pure-blood" Momenta R7 program, except for the MG 07, were all positioned at level 300,000 or higher. Therefore, in his view, this intelligent driving solution must give customers an "unexpected" experience. Cao Xudong (left), CEO of Momenta, and Chen Zhu (right), General Manager of SAIC MG Brand Business Unit, communicated with the core media. Cao Xudong even believed that their comparison was not the Xiaopeng Mona M03 with outstanding intellectual driving ability in the current 15-level sedan, but directly called the Tesla FSD V14 version, and even called the MG 07 "Tesla for young Chinese". How strong is the Tesla FSD V14 version? According to Cao Xudong, a year ago, when FSD was still V13, he went to Silicon Valley to investigate and found that less than 40% of users used or subscribed to this feature. After the FSD was upgraded to the V14 version, we went to Silicon Valley to investigate, and the subscription rate of new car buyers has become 100%. In Cao Xudong's view, the competition for intelligent driving in the future will be mainly between China and the United States. According to his judgment, "China's first place must be the first place in the world." Not only in the Chinese market to benchmark against Tesla FSD, Momenta even went to the European market to fight with Tesla. Cao Xudong expects that overseas markets such as Europe and the Middle East will liberalize Noa regulations from 2027 to 2028. They are ready to seize the European market in 2028. Cao Xudong's strength mainly comes from data. At present, Momenta has obtained more than 10 billion kilometers of driving data and extracted more than 100 million long-tail scenarios. The data comes from the 800,000 vehicles that have been equipped with the company's smart driving products. It is expected that by the end of this year, this data should reach about 2 million units, surpassing the number of Tesla FSD users. By the end of 2028, Momenta expects to have more than 10 million vehicles equipped with its own products. Data is not only quantitative, but also qualitative. Cao Xudong said that due to the more complex road conditions, China's intelligent driving data is 5-10 times more efficient than the United States, and some extreme scenarios are even 10-100 times more efficient. In his opinion, Tesla is an old driver in the United States to go to Europe, and Momenta is an old driver in China to go to Europe. In order to further consolidate the advantages, Cao Xudong revealed that Momenta will make a threefold increase in computing power investment in 2026. He expects to reach 60% of Tesla's total investment in North America within the year, and may even increase it to 60% to 80%. This scale is also at the forefront of the industry. Hashrate investment not only depends on smashing money, but more importantly, how to efficiently make kcal clusters and kcal clusters, to quickly train large models, and quickly iterate. Cao Xudong said that the post-training iteration of the Momenta R7 model only takes 2 to 3 hours at a time. After the pre-training optimization is completed, iterate once every two or three days, so that users can have a new experience every week. Not only does the intellectual driving ability show advantages among peer competitors, Chen said that the MG 07 also has two major competitive advantages, one is the "face value" of the veil that has not been fully unveiled, and the other is the smart cockpit that will be released in the future. Recently, the road test spy photos of MG 07 flowed out. Showcasing the new car's low profile and wide body posture, smooth slip back lines, front and rear light belts, and sporty shovel-shaped front lip shape, the car also features a frameless door and hatchback rear door design. According to Chen Shui, MG 07 is equipped with China's most powerful language model in terms of user size. Obviously, both the MG brand and Momenta have high hopes for the MG 07. It also makes the expectations of the MG brand's most important product in 2026 full. Is this new car really a hit, or is the gimmick bigger than it really is? It is expected that after 2 months, it will be officially known. The following is the original text of the exchange (compiled by Car House): Tesla for young Chinese? Media: General Manager Zi, MG 07 has recently received a particularly high level of attention on major social platforms. What new changes will the smart electric vehicle industry bring after the new car comes on the market? Chanseki: MG 07 only announced the naming and styling details in March. Many users pay attention to LiDAR after seeing the car. In the future, MG 07 will adopt the solution of MomentaR7 to strengthen the world model + X7 to develop a special chip for large models, and use the software and hardware integration solution to meet user needs. Styling is our know-how. 100 years ago, MG made luxury cars and coupes. We also see that there are many models on the market today that users want to look better, and we also see that the 200,000-class influencer coupes sell particularly well. For us, we are a brand that has built coupe for 100 years, why don't we build a car like this? Based on our confidence in the accumulation of coupe making, we are confident that a better coupe will come out and play our advantageous long board. In terms of the market, coupe buyers are younger and more personable. Intelligent driving is already a necessity for this population. So, when we built this product, we made intelligent driving a very high priority. Throughout the program selection process, our only option was Momenta. Because it has very strong capabilities with Momenta itself, and also has a good foundation for cooperation with SAIC. Due to the deep cooperation between the two sides, we were able to see the advantages of the R7 in advance, so we wanted to put this package on this car. I am very confident in the MG07 because it is very, very top of the line, both in terms of form factor and smart driving. Media: In the industry, intelligent driving algorithms now have world models and VLA routes in parallel. Everyone will be curious, why does the R7 model on the MG 07 choose this route, and is it considered to be a tri-service meeting with Xiaopeng VLA and Huawei ADS4.0? What are our core differentiators? Cao Xudong: I talked about similar issues in a media exchange two days ago. Later, a friend sent me a hot post on Xiaohongshu. Below is a comment with a particularly high response volume: A company that eats VLA is not as good as a version. Remove L Xiaopeng and make rapid progress. Xiaopeng's VLA actually has no L, it is a VA model, so it is essentially an end-to-end reinforcement learning model. This algorithm was also implemented on the R6 Reinforcement Learning Big Model last year. This year, we implemented a world model of reinforcement learning, benchmarked against the Tesla FSD V14. We hope that the MG car will become a Tesla for young Chinese. A year ago, when FSD was still V13, I went to Silicon Valley to do research. Less than 40% of people used or subscribed to Tesla FSD, which is already a very high proportion and exceeds the average. Recently, after the release of Tesla FSD V14, I went to Silicon Valley to do research. 100% of the people who bought Tesla subscribed to FSD V14. I also hope that, together with MG, we can build a young Tesla in China, and our intelligent driving is a Tesla FSD V14-level experience. Chen Chui: When we held the MG4 conference in 2026, we said that we had broken Tesla's monopoly. Throughout the industry, only Tesla has the technical ability to solve thermal management problems. But on the MG4, we also used an all-in-one thermal management system and broke down Tesla's patent barriers with an aluminum rare earth alloy material that was pioneered in the industry. Now, we have a very good reputation for power consumption and thermal management of our products. I listened to President Cao's suggestion. I understand the intellectual driving level of the MG07 car, and I can also break my wrist with the Tesla V14. Media: Your review of Xiaopeng VLA just now was OK. Nowadays, the MG 07 is positioned at about 150,000, which is a little competitive with the Xiaopeng hot-selling model Mona M03. The Mona M03 is equipped with a distilled version of the VLA. What is the difference between VLA2.0 and VLA Distillation? Based on this distinction, does Cao always think that R7 is the leading level in the 150,000 yuan product? Cao Xudong: Distillation is a very mature technology, or a very common technology since deep learning. More than ten years ago, before the founding of Momenta, when the performance of the mobile phone was not so strong, I told my partners that in order to run the model on the transfer machine, first train a larger model, and then distill a small model to run on the transfer machine. The performance of the distillation model depends entirely on how big the difference between the big model and the small model is. If the difference is small, it is possible to achieve 80% to 90% performance proximity. If the model size varies greatly, it may be 50% to 60%. Frankly, Xiaopeng is progressing fast, but not our main target right now. What we want to build is a Tesla for young people in China. So our product experience is targeted at Tesla's FSD V14, both in China and globally. Media: Data volume is a nourishment for the growth of intelligent driving algorithms, what is the data volume of the R7 model? If you run the world model on the end side, what are the requirements for the computing power and performance of the chip? Cao Xudong: We now have more than 10 billion kilometers of driving data. At the same time, more than 100 million long-tail scenarios of high-quality gold data were extracted from tens of billions of kilometers to train our large models. How many cars does this data come from? There are now more than 800,000 cars, and it is expected that there will be about 2 million cars by the end of this year. Tesla now has about 1.1 million total FSD subscriptions. When multiple peers give a forecast, they will reach a hundred million by the end of the year. In other words, by the end of the year, our cumulative city Noa could overtake Tesla. In terms of data, we are growing very fast and the whole flywheel effect has already started. After all, Tesla is just a car company, and we have multiple partners, and all the data can be aggregated into a data flywheel, accelerating, accelerating and accelerating. Unlike ordinary general-purpose chips, our chips are designed for large models. We have a strategic judgment from the beginning of 2022. Although it was a cold winter chip at that time, because GPT came out, the entire development direction of artificial intelligence will converge to a large model. If we customize and optimize the chip according to the structure of the convergent large model, there may be a 5-10 times performance improvement. This chip on the MG car has the calibration power of a Nvidia Thor, and its performance is very strong. Media: What is the profound significance of Momenta running on the road over Tesla FSD? Cao Xudong: Tesla's cars are mainly in North America, and ours are mainly in China. According to previous research by car companies, China's data efficiency is 5-10 times that of the United States. For example, a pedestrian crossing on the road may be encountered once per kilometer in urban China, but in the United States, it is at least 10 kilometers on average. Two-wheeled vehicles traverse, the probability of China will be higher, China has a very powerful takeaway and courier platform, so electric vehicles traverse is very, very much. I have a customer from abroad who said that he particularly likes to drive in China because the Chinese scene is very complex and challenging, like playing games. In contrast, driving in the US or Europe can be relatively boring, with few unpredictable events occurring. And in China, there will be a lot of unpredictable events, a bit like a clearance game. The efficiency of Chinese data is 5 to 10 times higher than that of the United States, or refers to some common scenarios. The limit scenario gap may reach 10-100 times. That's why we're going to Europe to benchmark the Tesla FSD V14. Tesla is the old driver of the United States to Europe, and we are the old driver of China to Europe. Who do you think will win? Media: Ji Zong, I feel that you are more confident in the intelligent driving of the MG 07. But Momenta doesn't just serve MG as a brand. In your opinion, where does the MG 07 differ in terms of intellectual driving? Chen Cui: Intelligent driving is particularly important for the customer base we want to target next, so now choose the best partner to do intelligent driving, do different models, and play my advantage in this segment. Because the customer needs and likes it more. The MG07's Momenta solution is a combination of hardware and software. Just now, many teachers mentioned Tesla and Xiaopeng LVA 2.0, including ads. Now in this market, the head of the intelligent driving first echelon is on the track. Everyone uses a software and hardware solution, at least Tesla is. We said at the launch that we were among the first to use Momenta, a software- and hardware-integrated reinforcement learning world model. To the best of my knowledge, this package is currently only available on 300,000 + products in addition to ours. So, in this segment, this set of smart driving must give customers an experience that exceeds their expectations. # 1 in China = # 1 in the world? Media: What does the collaboration between President Cao, Momenta and the MG brand mean for China's smart driving industry? How do you see the trend of China's AI out of the world? Cao Xudong: The SAIC Group has done a very good job in going out to sea, and it has already explored the territory before the epidemic. Before the pandemic, in 2018 and 2019, overseas sales were 100,000 and 200,000 per year. After the pandemic, the annual sales volume reached more than 1 million. The MG brand originally had a large accumulation in overseas countries such as the United Kingdom and the United States. Therefore, in terms of going to sea, we are the vanguard of China's going to sea. In terms of smart going overseas, we have more or less shared the overseas dividends of SAIC and MG brands. We judge that China's intelligent driving is about 3 to 4 years ahead of the world. In 2024, it may be the inflection point of intelligent driving technology in the Chinese automobile market. In 2028, overseas cities Noa or urban intelligent assistance may usher in an inflection point. We learned from various channels that the time point for the successive opening of Noa regulations in overseas cities is almost 2027. Regardless of Europe, the Middle East, and Southeast Asia, the time for countries to liberalize is probably at this point. Therefore, many OEMs are already targeting mass production Noa in 2027 and 2028. We set up offices overseas very early. Overseas offices now include the headquarters of three main engine plants in Stuttgart, Munich and the Middle East, Abu Dhabi, and Japan. We have experience in mass production and development overseas. We believe that with MG's existing accumulation and channels around the world, we will be the first player in the world to be able to mass-produce Noa in overseas benchmarking time, and we will definitely be able to benchmark the best player in the world - Tesla FSD V14, and in some ways even better than Tesla FSD V14. Media: Mr. Chen, MG, as the vanguard of China going to sea, why did you choose to cooperate with Momenta? Chen Chui: Cao is always a very low-key person. Next, we will present the Momenta R7 intelligent driving solution at the MG 07 meeting, including a complete set of software and hardware solutions for self-developed chips. Today, eight of the world's top 10 automotive groups use Momenta. Momenta has achieved a 61% share of third-party intelligent driving, ranking first in the industry for three consecutive years. BMW, Mercedes, Audi, the global benchmark for luxury brands, have either reached a strategic partnership with Momenta or are planning to use Momenta's solutions in the future. MG is a global brand. Throughout the EU plus the UK market, MG has sold more than 1 million vehicles since it became part of SAIC. It is a very iconic milestone for the entire Chinese brand to go to sea. SAIC is Momenta's largest institutional investor and is willing to stand with Momenta because Momenta's smart driving solution is strong enough and has great potential. Based on our cooperation with Momenta, we also want to stand on the world stage and let more Chinese people go to sea for China's intelligent driving and AI, so that the global media and users can re-recognize China's strength. Previously, we held a technical conference in Frankfurt, Germany, to showcase our semi-solid-state battery technology, and 223 mainstream media around the world covered the whole page. At that time, I really felt very proud in my heart. We also hope that MG and Momenta will put a great smart driving solution on the global stage in the future, so that more people can know MG and know Momenta. The entire collaboration between MG and Momenta is a new model of collaboration between international brands and technology companies. We unify goals, empower each other, and innovate together. The entire intelligent driving solution is the result of everyone's integration, constantly polishing the algorithm, the onboarding experience, the end-user experience, and constantly correcting. This is a collaborative model that we endorse at our roots. Media: Just mentioned that we have been talking about intelligent driving for half a day, in fact, we also need a good cockpit now. Because two days ago, Youshang released the cockpit integration, I don't know if MG's vehicle will have any cockpit extension? Chen Shui: Regarding the smart cockpit technology of this car, it has not yet reached the communication node. Last year, we were already doing a lot of layouts throughout the smart cabin experience. As you can see on our previous posts, we and the 3C head brand have created an intelligent experience for the entire cockpit vehicle. Today, the smart experience on your phone is very convenient and can meet all your needs. It is possible that all of today's smart devices will not feel smarter than the phone you have in your hand. The smart cabin experience in our car will be more and more like a mobile phone, giving you the feeling of a smartphone, including how smart it is. That's one thing we're doing. Relying on SAIC's entire circle of friends, including the 3C brands, including Momenta, we have selected the top solution partners in the business field to do it. I actually just heard today that we are going to benchmark against Tesla FSD V14, and we are going to be Tesla in China. This is quite exciting for me. Cao Xudong: The entire SAIC Group has a very large depth, whether it is a variety of brands, no matter what type of car, for the entire intellectual cabin and intelligent driving attention is extremely high. The one thing I can reveal is that the large language model onboard MG07 is definitely one of the only large language models with a head in China. It also has the best intelligent driving model in China. Chen Chui: President Cao said that China is the only one. In fact, from the perspective of user use and coverage, it is the only best large language model in China. AI solutions, which will be seen next on the MG brand. Media: You mentioned Benchmark Tesla, Huawei, these are very good companies. How do you feel about the progress of these companies in terms of technology? Cao Xudong: Tesla has a particularly strong Silicon Valley gene in the United States. In terms of technological innovation, I have a peer evaluation that is particularly good. He said: Tesla's research and development is only upgraded, not iterated. He is basically upgrading the system and architecture, rather than doing this thing through iteration under the same set of architectures and systems. China's relatively mature large companies tend to be more iterative, pushing the relatively mature architecture system to the extreme through saturated organization and resource investment. These are two very different styles with advantages. But in the rapidly changing era of AI, Tesla is the more correct model. We are always in a state of competition, creating the most advanced technology through you chasing me. I think the competition of intelligent driving is the competition of China and the United States. And I judge that the first place in China must be the first place in the world. “New experiences every week” media: There are now purely visual routes, as well as lidar and camera routes. Momenta is more adhering to the route behind, what is the logical thinking behind it? Cao Xudong: The entire solution of the sensor, we mainly visual, lidar can also be used. It will be better after use. In some scenarios, such as tunnels, if an accident occurs at the tunnel entrance, the visual response time will be longer than the lidar. Vision solves this problem, too, but with lasers it's better. Our entire architecture is designed with vision in mind, but lasers are optional. Choosing a laser will also bring some performance improvements. Some of our customers, such as Mercedes-Benz and BMW, choose the purely visual technology route. Of course, domestic customers also choose the technical route of lidar, which will be reflected to a very high level. Intelligent driving ability lies first and foremost in data and algorithms, which play the most important role. The second stage is the computing power, especially the ability to increase the computing power of large models and chips by 5 to 10 times; the third stage is the sensor. As long as the sensors are enough, most of the more sensors are icing on the cake, and it is difficult to rely on the sensors to bring 100 times or 1000 times improvement. Media: A friend recently announced that the investment in computing power has reached 10 billion, and the future will reach 70 billion to 80 billion. What kind of layout does Momenta have here? Cao Xudong: The input of computing power is very, very important. We have tripled our hashrate investment this year compared to last year, and it is inconvenient to disclose the exact number when it comes to trade secrets. But I believe that our computing power level may reach 60% of Tesla's total investment in North America this year, and may increase to 60% to 80% of Tesla's FSD training in North America. Therefore, the scale is quite large, and it is absolutely one of the best in the industry. Nowadays, investing in computing power is not just about spelling and throwing money, because everyone can do it. And more importantly, after having these GPUs and machines, how to become a kcal cluster and a kcal cluster, to quickly train large models, and quickly iterate. This is the more fundamental purpose of a company. Spend money everyone will. But after spending these funds, who's R&D investment is 1, and who's R&D investment ROI is 10, there is a big gap. For example, our current large model training, our post-training iteration only takes 2 to 3 hours at a time, MoE train (mixed expert model training) can be iterated once a day, after pre-training optimization, iterate once every two or three days. It is precisely because of this speed of iteration that we can achieve a new experience every week on the R7. So with the MG 07 on our R7, the smart driving experience is something to look forward to. Media: Did Momenta predict what the pace of adoption of physical AI would be like? Cao Xudong: The first large-scale explosion of physical AI is autonomous driving. The biggest challenge in other application areas of physical AI, such as robotics, is that the ontology is not converging. OpenAI originally wanted to be a robot, but ultimately focused on digital AI. Because what exactly is the body of the robot? How to design? Whether it is a completely human-like ontology or a different ontology design for different vertical applications, the entire industry has not converged at all. And the great challenge brought by no convergence is that the ontology is not well trained, and the data is naturally not well trained. There is no amount of data, the entire flywheel cannot be rotated, and it may still be in a very preliminary stage. Large model training requires a large amount of data. For example, digital AI is basically not enough to eat the entire Internet data, and it is not enough to take all the library books and sweep them into the training. Everyone tried every means to put all the data in, still not enough to feed this big model. So, the big model is actually very data hungry. Now, the amount of ontology data of the robot is far from feeding the large model of 1T. This is a big problem encountered by embodied intelligence: the ontology has not converged, the commercial closed loop has not formed, and the data is very lacking. Autonomous driving is a pearl that goes to the forefront. Because its data volume has been scaled up, a commercial closed loop has been formed. Now we have 800,000 cars and expect to have more than 10 million by the end of 2028. This amount of data is massive. Data based on autonomous driving enables World Model training. The improvement is enormous. Media: There is Moore's Law in the field of electronic technology, is there also Moore's Law in the field of intelligent driving? Cao Xudong: The Moore's law of intelligent driving is that it increases by 10 times every two years, but it is accelerating at the moment. For example, Tesla and us, we are confident that we will be the leader in the industry and even achieve 10 times increase every year. The experience enhancement and business model transformation that this brings is enormous. As mentioned before, I did research in Silicon Valley. A year ago, Tesla's FSD subscription was less than 40%, and now the small sample survey is close to 100%. That's why Elon Musk dared to launch a subscription-only model. The subscription model is more challenging because users only pay for one month, and if it is not easy to use, they will not pay afterwards. Unless the product is good enough and the renewal rate is very high, he will not dare to choose a 100% subscription method. Media: How does physical AI empowerment help Momenta's 10-year vision? Cao Xudong: It will definitely accelerate. Our ten-year vision is: ten years, saving millions of lives; ten years, liberating 100% of time; ten years, doubling the efficiency of logistics. In essence, it depends on the fact that the safety of autonomous driving is ten times, a hundred times, and a thousand times better than that of humans, so that it can be realized. Now, Moore's Law is accelerating from 10 times in two years to 10 times a year. It is believed that the three “ten-year visions” will be achieved in a large proportion by 2030, and may even be 100% achieved or over-fulfilled.