On September 22-24, the 2026 World New Energy Vehicle Conference was successfully held in Haikou, Hainan. The conference was jointly hosted by the China Science and Technology Association, the Hainan Provincial People's Government, the Ministry of Science and Technology, and the Ministry of Industry and Information Technology. With the theme of "transformation, upgrading, and resilient growth", the conference brought together global vehicle giants, core supply chain enterprises, and cutting-edge scientific and technological forces to jointly explore the development path and new paradigm of international cooperation for new energy vehicles in the context of low carbon and intelligentization. Chen Liming believes that Noa in urban areas is changing from the highlights of past car purchases to a must-have option for models above 200,000 yuan. The industry is also returning to the product experience and safety bottom line from L2.9-type over-marketing. This year, we should pay more attention to the real indicators such as the opening rate of intelligent driving, the number of takeovers and the intelligent driving mileage ratio. He said that intelligent driving brings functional value, intelligent cockpit brings emotional value, and the horizon is pushing the car to a super-agent around "functional experience + emotional experience". He pointed out that HSD, as the first pure one-stage end-to-end large model in China, has been mass-produced in Xingtu ET5, Deep Blue L06 and other models. Among the 100,000-20,000 models, more than 60% of users choose the high configuration with Noa in urban areas. The monthly use mileage of HSD bicycles has increased from 171 kilometers to 573 kilometers, indicating that users are willing to pay for safer, smoother, and more like older drivers. He also said that one-stage end-to-end, world model and reinforcement learning are the key to performance improvement, and HSD 2.0 has achieved the ability to turn around narrowly and multipointly. Starcraft fusion chips do not pursue the highest single computing power, but rely on deep soft and hard collaboration to reduce the number of chips by 50%, reduce R&D investment by 70%, and save the overall cost of the vehicle by 1500—4000 yuan, laying the foundation for the popularization of intelligent driving. Horizon Vice Chairman Chen Liming gave the following speech: I am very honored to have the opportunity to report and share Horizon's innovations and achievements in user-oriented, user value-driven, collaborative innovation over the past year to make the vehicle a super-agent. Let's take a look at the situation of the smart driving market. In the first half of this year, the sales volume of passenger cars decreased by 20% compared with the same period of last year. The Noa of urban areas grew against the trend. The penetration rate of Noa of self-owned brands in urban areas has reached nearly 30%, an increase of 15% compared with the same period of last year. The Noa of new energy in urban areas has reached 40%, an increase of nearly 20%. For the 100,000 and 200,000 + cars, this increase was even higher, at around 24%, reaching 52.8% and 88% respectively. It can be seen from the penetration rate that the Noa of the car city area of more than 200,000 is becoming the first choice for users to travel. McKinsey has a survey of Chinese consumers. In the past three years, the choice of urban Noa has become a must from the fifth place in the past to the third place in last year, and this year's urban Noa of more than 200,000 has become a must. We take an in-depth look at the changes in the cognition of Noa users in urban areas throughout the market. 2024 is the burst point of Noa marketing in urban areas. Most of us listen to L2.9, L2.999, etc., which can be opened everywhere in the country. These excessive marketing has also produced some misleading and more serious vicious incidents for consumers. Relevant state ministries and commissions promptly issued some policies, standards and regulations to rectify and regulate the market, clarify the safety bottom line, and clarify the functional boundaries, so that consumers have a deeper understanding of safety and intelligent driving performance. With the increase in awareness, this year is more about the experience of returning to products, such as using the opening rate of intelligent driving, the number of mileage takeovers between malignant events, and the intelligent driving mileage ratio to judge the performance evaluation and use of the intelligent driving system. The product experience can be refined into two aspects: the functional experience and the emotional experience. The main function is easy to use and reliable, and the emotional experience is love to use, rest assured, and it is very pleasant to use. Intelligence has two carriers: smart driving and smart cockpit. We believe that intelligent driving brings functional value, that is, functional experience. The intelligent cockpit brings more emotional value, that is, emotional experience. Below, I report the results of the horizon over the past year around these two aspects. For the functional experience, there is a product called Horizon Super Drive, which is an L2 advanced assisted driving system. It was officially launched in November last year. It is the first pure one-stage end-to-end large model in China, mainly to provide customers with a safer, smoother and more like an old driver driving experience. This system was first released in mass production on the Chery Xingtu ET5 and Chang 'an Dark Blue L06. Multi-configuration models with a price of 100,000-200,000, what proportion of customers choose the high-configuration with urban Noa? The industry average is 18.7%, less than 20%. According to the sales data of Xingtu ET5 and iCARV27 from January to July this year, more than 60% of users have chosen to have a high configuration, with HSD, which is three times higher than the industry average. In this regard, the main thing is that Horizon HSD does provide a safer, smoother, more like an old driver experience through user use and media evaluation. Users are willing to pay for a good system with real gold and silver, and one of the important reasons why the horizon system can provide a safe, silky, old driver experience is innovation. Another data worth mentioning is the (HSD) single-month use mileage of bicycles. HSD1.0 was released in November last year. In January, the single-month use mileage of bicycles was only 171 kilometers. In June this year, HSD2.0 was released. In August, we made statistics. The single-month use mileage of bicycles increased to 573 kilometers, an approximate increase of 3.4 times. In fact, behind it is the user's trust in the system, it is easy to use, reliable and more use it. What is the logic behind what I just mentioned? Let's start by sharing some of the innovations that we've been working on. Like the CTO of Nissan just mentioned, the performance architecture of the system has been iterated, like the rules we just talked about to introduce neural networks, step by step to replace the past rule-based framework, and finally developed into a paragraph, which is divided into portrait, horizontal and parking. At present, the industry mostly uses the framework of a main model plus a sub-model. The idea of the horizon, we know that when people drive straight, it is an idea. When they turn horizontally, it is another system, or when they reverse, they have to switch. It is very natural. It is a model that drives and parks in the brain. Here we are purely piecemeal, portrait, landscape and parking, all the driving models are integrated in one model, so that one model can achieve the driving task. This model is to achieve unlimited startup of the whole scene, and there is no mode switching, because no mode switching will not feel the breakpoint during the entire driving process, which is why it is silky. In addition, in terms of model innovation and advancement, 90% of the training data is road real data. There are also some long tails and corner cases, we introduce world models and reinforcement training, and through advanced models plus world models and reinforcement learning, the overall performance is greatly improved. At the same time, we can see that the potential of purely one-step large models can be further tapped and emerged, and I will share many specific examples later. The overall reaction speed has increased by 20%, the parking success rate has increased by 50%, the safety has improved by 56%, and the game ability has improved even more (167% improvement). Specifically, let's take a look at the specific aspects of performance improvement, which are mainly reflected in six dimensions: one is the overall upgrade of the model, and at the same time, it is a security upgrade, which was also mentioned earlier. There are also upgrades to driving and parking, as well as environmental rendering, which ultimately leads to an upgrade to the user experience. In addition to innovating in high-tech, Horizon is also very concerned about safety. In terms of safety, there have been major improvements in emergency auxiliary braking, emergency avoidance, and accidental step acceleration pedal prevention. In terms of driving, the industry has a safety headache or a difficult problem to solve, that is, a turnaround on a narrow road, especially a multi-point turnaround. We have also achieved a multi-point turnaround in the new HSD2.0. At the same time, it also mentions roadside in-situ activation and roadside parking recommendation, which is actually an L4 function, which is realized in this part of our system, which is also the function and capability realized in the purely sectional large model. There are also upgrades to environment rendering. In addition to driving functions, we also provide environment rendering for human-computer interaction, so that you can see the surrounding environment, other vehicles on the road, vehicles, pedestrians, etc. in real time. Users can choose the characteristics of surrounding vehicles according to their preferences. For example, you like to drive sports cars and sports cars together around you. You can also choose to drive SUVs around you, or cars and RVs, or even cars with horses, animals, cats, dogs, etc., according to your needs. In addition to providing a better driving experience, HSD also provides some emotional value, which is an upgrade to HSD functionality. Functional upgrades result from the accumulation of continuous innovation and the translation of innovation into specific products. Using the world model improves the overall system capabilities and further explores the cutting-edge of the world model. One of our papers, Epona, was selected by ICCV2025. In the video, you can see that it generates a longer video from the image through the video, so that simulation training can be carried out. The Vista and Cosmos on the left are both industry benchmarking methods and models. After a few seconds, the models become blurred and there is no data available. Epona also has a good driving scene after 20 seconds, which helps us train the model well, especially the long tail phenomenon. Corner case can be simulated to further provide data for the large model and further improve the ability of the large model. This is some work on the driving side. We have also seen a current trend, that is, the entire electronic and electrical architecture has evolved from a distributed domain processing architecture to a centralized one. Horizon also released another generation of new series of chips called Starry Sky during this year's Auto Show. The original journey was mainly for intelligent driving. The Starry Sky chip series is a cockpit fusion chip. It is the first cockpit fusion chip with large computing power and the most advanced process in China. It is a 5-nanometer process. AI has 650tops, GPU has 3000GFLOPS, and CPU has 500k DMIPS. At the same time it comes with 6 screens for a very smooth experience. Here, you can see the rendering and smoothness of 3D cockpit images, as well as 1.4 times that of the industry's best cockpit chips. In terms of autonomous driving, the Transformer also handles nearly 10 times faster than the most powerful chips on the market. It is also 2.4 times faster than the best chips on the market today in terms of preprocessing speed for large end-side models. You can see that whether it is AI computing power or GPU computing power and CPU computing power, it is not the best and highest compared to other chips in the industry, but it can provide better performance, and the most important thing is depth-level hardware and software innovation. Through deep software and hardware collaboration, better hardware performance can be stimulated to handle the corresponding functions. Here we see a performance boost. At the same time, because the cabin cockpit fusion turns two chips into one chip, the number of chips is reduced by 50%, and the two domain controllers become one. At the same time, R&D investment was reduced by 70%. In addition, in the current local shortage (storage), from the original storage requirement of 48GB to 28GB, the demand for memory has been greatly saved. The comprehensive cost of the vehicle can save 1500-4000 yuan, which is a very good basis for popularizing autonomous driving to all users. What I just mentioned is the performance improvement. Next, I would like to share the new product we released in April this year in terms of emotional experience - coffee shrimp, which is a vehicle intelligent operating system. Coffee shrimp can call both physical and digital agents and large models in the cloud at the same time, providing smoother human-computer interaction. It has three characteristics: more personality, smarter understanding of you, can remember, like a smarter, more understanding of your intelligent partner, so that the entire cockpit intelligent experience from the original people feel special machinery, with one do not want to use the interaction with the machine, into a very natural, more understanding of your intelligent partner. The horizon is more above the iceberg, in fact, below the iceberg, the accumulation of long-term investment and research and development of the horizon for 11 years. In the past three years, we have published more than 50 papers at international top-level conferences, with more than 1 billion citations. Especially in CVPR2026, 11 papers were selected, and more than 2,000 patent applications were filed worldwide. Many of the driving innovations also come from the horizon. I will not introduce them one by one here due to time. The car is the first carrier of physical AI and is becoming a super-agent. Horizon is also providing the underlying technology platform and infrastructure for cars to move towards super-agents. In terms of smart cockpits and intelligent driving, we have the Journey series and the Starry Sky series, and in terms of software, HSD and coffee shrimp, which provide a solid base for creating super-intelligent platforms. Due to Horizon's innovation over the years and customer recognition of the quality and performance of our products, Horizon continues to lead the market in terms of smart chips to date. Achieved second place in urban Noa and we hope to have better results next year. At the same time, we have also been recognized by more than 40 OEMs. At present, we have more than 500 fixed-point models, of which 350 models have been mass-produced and put on the market to serve users. Recently, we just celebrated the shipment of 15 million chips, so finally we also want to work with our colleagues in the industry to drive our innovation with customer value, drive our growth, and work with the industry to make the car a super-agent. This is the end of my sharing, thank you!.