GuShiio.com ① The full-blood version of DeepSeek V3/R1 was launched on Huawei Cloud, and the agency said that AI model iteration is expected to accelerate;② Xiaopeng Automobile will launch MONA M03 MAX in May, equipped with the industry’s top autonomous driving technology;③ The Byte Bean Bag Large Model Team proposes a sparse model architecture, which can reduce reasoning costs by up to 83%.
[Today’s Introduction]
DeepSeek V3/R1 full blood version launches Huawei Cloud, agencies say AI model iteration is expected to accelerate
Xiaopeng Automobile will launch MONA M03 MAX in May, equipped with the industry’s top autonomous driving technology
The Byte Bean Bag Big Model team proposed a sparse model architecture, which can reduce reasoning costs by up to 83%
Guo Minghao: Apple is studying humanoid and non-humanoid robots
This big model is about to be released, and the industry is ushering in important opportunities for intelligent upgrades
BYD’s 70,000-class high-end smart driving cars are officially launched, and institutions expect China to usher in an inflection point for smart driving this year and next
In 2025, eVTOL is expected to gradually transition to semi-solid state, officially opening up the low-altitude market for solid-state batteries
[Theme Details]
DeepSeek V3/R1 full blood version launches Huawei Cloud, agencies say AI model iteration is expected to accelerate
According to Huawei Cloud WeChat information, the full-blood version of DeepSeek V3/R1 has been launched on Huawei Cloud. DeepSeek V3/R1 671B flagship model (full blood version) usually requires high-end GPUs for reasoning. Now, based on the full-stack optimization and adaptation of Huawei Cloud Tengteng Cloud Services, it can achieve the same effect as the global high-end GPU deployment model and provide stable, Production-level service capabilities to meet business commercial deployment needs. Huawei Cloud Shengteng Cloud service can provide surging, flexible, and sufficient AI computing power, and supports code-free, one-click deployment of the flagship model.
Shanghai Securities believes that the in-depth cooperation and nesting between IT companies and DeepSeek will have a significant positive boost to the model’s promotion of cloud service experience. At the same time, the widespread application of open source strategies brought by DeepSeek has changed the business model of closed-source giants, and traditional developers The collaborative approach has changed, and AI model iteration is expected to accelerate.
On the company’s side, the Deep Sea Model and the Baize Government Model launched by Nanwei Software have been connected to relevant versions of DeepSeek and privatized and deployed. Through deep integration of advanced models such as DeepSeek-R1 and DeepSeek-V3, model distillation technology is used to optimize model performance. Wanxing Technology has completed the in-depth adaptation of DeepSeek’s latest reasoning model DeepSeek-R1, covering multiple products in its video creativity, drawing creativity and document creativity software businesses.
Xiaopeng Automobile will launch MONA M03 MAX in May, equipped with the industry’s top autonomous driving technology
According to media reports, the nano-AI search “Hundred Vehicles Action” was held in Beijing. He Xiaopeng, chairman of Xiaopeng Motors, said that he would make a top smart driving product worth US$20,000. Xiaopeng Automobile will launch MONA M03 MAX, the top-of-the-line version of MONA in May this year,”equipped with the industry’s top autonomous driving technology.” MONA’s name comes from “made of new AI”. He believes that in the next 10 years, the penetration rate of high-end intelligent driving will increase from 2% to 50%, or even 80%-90%.
Minsheng Securities said that end-to-end technology promotes the leap in smart driving capabilities, with equal rights in smart driving as the core line. Car companies represented by Xiaopeng and Huawei have continued to iterate and promote smart driving since 2024. Xiaopeng launched the first mass production of an “end-to-end” large model in China. In addition, various manufacturers are actively promoting the popularization and promotion of smart driving, and the era of smart driving has arrived. Looking forward to 2025, the advancement of smart driving technology is expected to lower the threshold of smart driving hardware and apply smart driving in mainstream markets within 200,000. Car companies led by Xiaopeng and BYD are expected to form differentiated advantages in 2025 through affordable and intelligent development., and get an increase in sales.
On the company side, Huashi Navigation, as a leading domestic high-precision positioning company, has made breakthroughs in the autonomous driving business. Some models developed in cooperation with a number of car companies have achieved mass production and have now become the exclusive IMU-BOX for Xiaomi Automobile’s SU7 series products. supplier. The company can provide users with a complete set of laser radar products and related solutions. Desai Sev deeply focuses on the intelligent and efficient integration of automobiles, laying out smart cockpits, smart driving, and smart networking. The company’s intelligent driving domain control products rank first in the industry, with a market share of 23.90% from January to August 2024. The company’s product matrix will cover more computing power ranges to meet a wider range of market needs. The high-computing platform has mass-produced and supported many car companies such as Ideal Car, Xiaopeng Automobile, Lotus, GAC Aian, and Jikrypton Automobile.
The Byte Bean Bag Big Model team proposed a sparse model architecture, which can reduce reasoning costs by up to 83%
The Byte Bean Bag Big Model team proposed a sparse model architecture, which improves the reasoning speed by 2-6 times compared with the MoE architecture, and reduces the reasoning cost by up to 83%. In addition, according to the official announcement of the beanbag model team, the video generation experimental model “VideoWorld” was released. Unlike mainstream multimodal models such as Sora, DALL-E, and Midjourney, VideoWorld is the first time in the industry that it can recognize the world without relying on language models.
As a general video generation experimental model, VideoWorld removes the language model and achieves unified execution of understanding and reasoning tasks. This purely visual model can “predict” the future and “understand” causality. The computer team of Minsheng Securities believes that the release of “VideoWorld”, a large open source video model of bean bun, will make video generation a common knowledge learning method and act as an “artificial brain” in the real world, which is expected to activate the “visual market” to open up room for growth.
Among listed companies, Netda Software has always focused on the field of big videos. The “Blueprint” intelligent image creation system is applied to the image production process of content posters and content covers by using the cultural graphics capabilities of large models. Danghong Technology’s BlackEye multi-modal audio-visual model has completed deep integration with DeepSeek, and recently jointly launched the DeepSeek Media Industry Application Innovation Alliance.
Guo Minghao: Apple is studying humanoid and non-humanoid robots
Guo Minghao just posted on overseas platform X saying that Apple is exploring humanoid and non-humanoid robot projects for its future smart home ecosystem, but these internal products are still in the early proof-of-concept (POC) stage. Given the current progress and large-scale project development cycles, he believes that Apple robots will not be in mass production until 2028 or later.
At the same time, according to supply chain research, Apple is more concerned about how users establish interactive perception with robots rather than the appearance of robots, which means that sensing hardware and software are the core technologies. Everbright Securities Chen Jianing believes that reinforcement learning is a method of learning through trial and error, but traditionally this method is very time-consuming, while DeepSeek’s training method is low in cost, and the R1 reasoning accuracy rate in three-dimensional space reaches 87.6%. Cross-modal intention understanding F1 value is 91.2%. If it is deeply integrated with the robot field, it may promote the robot to evolve from “perceptual intelligence” to “cognitive intelligence” and accelerate its application in more vertical scenes.
Among listed companies, Lingyizhi has the supply capacity of core components such as controllers, reducers, bionic robotic arms, servo drives, and power control systems, as well as the ability to assemble complete machines. Now it has cooperated with Hanson, Zhiyuan, FigureAI and other world-renowned humanoid robot companies cooperate in research and development to provide customers with various core components. Zhixin Precision has industry-leading technologies such as machine vision, precision motion control, and universal software platform. Its main customer is Apple.
This big model is about to be released, and the industry is ushering in important opportunities for intelligent upgrades
Huawei’s Data Storage Public Account released information stating that it will release a Ruijin pathology model based on Huawei’s DCS AI solution. According to reports, the model will be released at the “2025 Medical Artificial Intelligence and Precision Diagnosis and Treatment Development Forum” held on February 18.
Pathological diagnosis is a diagnosis method based on image information and is known as the “gold standard” for disease diagnosis. At present, the development of my country’s pathology industry is faced with the characteristics of lack of pathologists, uneven distribution, and long culture cycles. “Automation, standardization, digitalization, and intelligence” have become an inevitable trend in the development of pathology. Jianghai Securities pointed out that the breakthrough development of AI is promoting the leap of medical diagnosis and treatment from traditional empirical medicine to data-driven precision medicine. Under the guidance of the “Healthy China 2030” strategy, the medical industry is ushering in important opportunities for intelligent upgrades.
Among listed companies, the cooperation between Ampiping and Huawei focuses on full-process digital solutions for pathology departments, combining Huawei’s storage and computing power technology to solve the problem of storage and efficient reading of large amount of pathology slice data to ensure diagnostic efficiency. Daan Gene said that in terms of AI-assisted diagnosis applications, the joint-stock company Yunkang Group has successfully introduced projects such as pathological DNA polyploid AI-assisted diagnosis, cervical liquid-based cell AI-assisted diagnosis, and chromosome AI analysis. The diagnosis efficiency has been greatly improved and the application effect continues to appear. Dean Diagnostics said that in 2024, the company’s subsidiary Medical Policy Technology will assist the provincial-level digital pathology service system in Hubei Province, and has signed contracts with more than 70 local hospitals, with a local pathology AI market share of approximately 70%; Dean’s own laboratory has provided more than 3 million AI pathology diagnosis reports using related products; related applications have been extended to the CRO field.
BYD’s 70,000-class high-end smart driving cars are officially launched, and institutions expect China to usher in an inflection point for smart driving this year and next
BYD Automotive Industry Co., Ltd. announced that the Seagull Smart Driving version has been officially launched, claiming to be a “70,000-class high-end Smart Driving Mainstream Scooter” with a price starting at 69,800 yuan.
Ping An Securities believes that BYD has a strong demonstration effect of smart driving, competitors must follow up quickly, and smart driving functions, especially high-speed NOA, are expected to spread rapidly. Although BYD is a follower from the perspective of the release of smart driving functions, BYD is an important promoter of the rapid increase in smart driving penetration in the entire industry. Everbright Securities believes that as BYD and other leading new energy vehicle companies successively carry high-speed smart driving with NOA and above functions, it is expected that the turning point of domestic smart driving 2025E-2026E is expected to approach, and 2027E-2028E is expected to usher in a period of rapid growth in the 2.0 era.
Among listed companies, Wanji Technology relies on the advantages of roadside data to build a full-stack self-developed vehicle-road cloud integrated intelligent driving solution, develop single-vehicle autonomous driving solutions, open up vehicle-road collaborative data fusion, and provide L4+ high-level intelligent driving solutions for vehicle-road collaborative. Desai Sev deeply focuses on the efficient integration of various fields of automotive intelligence, and its business covers smart cockpits and smart driving. Company customers include Geely Automobile, Great Wall Motor, GAC Passenger Cars, Changan Automobile, Chery Automobile, BYD, SAIC Passenger Cars, FAW Hongqi, Ideal Automobile, Xiaopeng Automobile, NIO Automobile, etc.
In 2025, eVTOL is expected to gradually transition to semi-solid state, officially opening up the low-altitude market for solid-state batteries
According to media reports, with the decisive layout of the eVTOL field of electric vertical takeoff and landing aircraft, power battery companies as suppliers are not willing to lag behind. Especially in the situation where the small-batch production of high-specific-energy solid-state battery soft-pack lithium batteries is more in line with eVTOL needs, the new layout and market development are expected to accelerate the mass production application of solid-state batteries and open up a new situation for power battery manufacturers.
Solid state batteries are hailed as the most promising next-generation power battery technology. Following consumer electronics, automotive batteries, and energy storage batteries, power battery companies have opened up a new track called eVTOL. Some market research institutions predict that by 2030, the global eVTOL market size will reach US$55 billion, the market size will exceed US$1 trillion in 2040, and the market size will reach US$9 trillion in 2050. The broad market demand, especially the accelerated layout of automobile companies in the eVTOL field, has directly driven a new race for power battery companies. Guolian Securities pointed out that with the first flight of Yihang’s smart solid-state battery, the cooperation between Ningde Times and Fengfei, and the low-altitude layout of other multiple solid-state manufacturers, eVTOL is expected to gradually transition to semi-solid state in 2025, officially opening the low-altitude market for solid-state batteries.
Among listed companies, Nandu Power said that combined with the current technical characteristics of the company’s solid-state batteries, the more matching application scenarios are low-altitude aircraft, drones, robots, etc., and is currently actively connecting with related companies for multi-scenario application and promotion. Shanghai Shiba is steadily advancing new solid electrolyte materials, solid/solid interfaces and battery devices, and will actively participate in the construction and demonstration applications of relevant industrial chains. The research and development of sulfide all-solid-state lithium batteries in the high-energy era invested by Lanhai Huateng is progressing smoothly. The construction of an automated production test line has been completed, and small-batch production of batteries within 5Ah has been achieved. Currently, preparations are being made for the construction of a batch production line.
Related stocksShanghai xiba-3.02%Huace Navigation+1.51%+2.39%Wanji technology+3.22%Zhixin Precision-3.61%Dean diagnostics+18.95%Lingyizao+1.87%danghong technology+5.05%lanhai huateng+1.81%wanxing technology-1.81%nanwei software+9.99%Nandu power-0.56%Wangda Software+4.87%Daan gene+10.07%An Biping+20.02%
