Log in to the Sina Finance APP and search for [Xinphi] to see more evaluation grades

During the Spring Festival of the Year of the Snake, the latest large model released by Deepseek stirred up the global AI market and stock market. Deepseek has shaken the industry with its low price, innovative algorithms and technology, and has also set off a wave of global capital to re-evaluate the value of Chinese technology companies.

After the holiday, the A-share market Deepseek concept stocks took off, and there were rumors in the market about Deepseek's related investors and partners, and many companies have recently spoken out through different channels. Panorama selects the most authentic voices from listed companies for you. (The following content is excerpted from Shenzhen Stock Exchange Interactive Easy, SSE e-Interactive, listed company announcements, etc.).

Do you have any products that have applications or cooperation in domestic large models?

TongTech[300379]: TongLMM, the company's large-scale model middleware, has been connected with domestic large-scale models such as DeepSeek, Tongyi Qianwen, Wenxin Yiyan, and ChatGLM. In the field of network security, the company has docked with Ali Tongyi large model in security orchestration, 4A and data security products, and realized many functional scenarios including network war room, 4A operation and maintenance customer service, vault semantic intelligent analysis, data classification and grading, etc.

Deepseek's high performance and reduction of training costs are the results of research and development at the national level, and China is overtaking in the AI field through algorithm innovation. Wanda Information is the leader in the industry with a large amount of data, how to lead?

Wanda Information [300168]: The company has organized a special team to conduct research on DeepSeek, and has completed the localized deployment and computing power integration of the open source version of DeepSeek in the company's own intelligent computing environment. It is used in industry scenarios such as medical insurance. The company focuses on new quality productivity, carries out in-depth innovation capacity building, and enhances the company's core competitiveness. First, focus on cutting-edge technologies and promote comprehensive innovation; the second is to keep up with market demand and lead product upgrading; The third is to strengthen R&D capacity building and improve the service level of the front line of business.

DeepSeek, which has recently become popular, has a low demand for computing power but can still be compared to OpenAI's latest products, so related concepts such as CPO and chips are affected. Your company is committed to the scheduling of computing power, and the future trend may have a greater demand for algorithms, how can your company flexibly respond to the new wave? Has your company cooperated with Deepseek?

Capital Online[300846] :D eepseek's strong rise in the field of artificial intelligence has positive significance for the cloud computing industry: 1. Based on the open source Deepseek, our cloud vendors can directly provide model services close to OpenAI's capabilities, which is conducive to the cloud computing industry to directly provide customers with mass services; 2. The open source and cost reduction of Deepseek will allow more enterprises and developers to enter the AI field, promote more industry models and AI application scenarios, and further stimulate the market demand for computing power, and the computing power industry is expected to usher in new growth opportunities with the expansion of AI applications. 3. Deepseek has successfully adapted to a variety of domestic hardware, and diversified chip deployment opportunities are coming, and more and more different chips will support models similar to Deepseek. The rise of Deepseek is a vivid demonstration of the scientific and technological strength of the whole of China. The company is proud to witness the glorious journey of this fellow traveler. With keen market insight and bold action, Capital Online has gained insight into the development trend of the industry as early as 23 years, and has formulated a business development layout of "one body and two wings", with the integration of intelligent computing cloud as one, one cloud with multiple cores and open computing pool as the two wings. One cloud and multiple cores are in line with the diversified chip deployment of Deepseek open source models; Key capacity building such as domestic card adaptation and model tuning is conducive to the selection of computing power solutions for more different models. It is based on these advantages, in the face of this opportunity, the company can make precise efforts, take advantage of the trend, and strive to create greater value for shareholders. Recently, Deepseek related concepts have continued to be popular, and the company calls on investors to pay attention to the company's medium and long-term development, pay attention to rational investment, and prevent concept speculation.

Yesterday afternoon, I tried the Westlake-Omni at a medical facility. It shocked me! The content and speed of the answers are far less than ChatGPT in 2022. Domestic leading companies such as Wenxin, Tongyi, Kimi, Zhipu, etc., not to mention Deepseek, have long surpassed the 2022 version of ChatGPT. If Tom Cat does release a physical doll, it is difficult to imagine that Westlake-Omni will be embedded for communication. In fact, half a year ago, the company said that it would sell physical dolls before the Spring Festival this year, but now it no longer mentions this matter, please reveal more truth.

Tom Cat [300459]: The AI intelligent hardware product Tom Cat AI voice emotional companion robot developed by the company has been launched in December 2024 on Tmall, JD.com, WeChat, Douyin and other major domestic online and offline channels, the product is equipped with the "Talking Tom Emotional Companion Vertical Model" and the speech synthesis model of West Lake Xinchen, and called some models in the bean bag series model, and is testing the capabilities of other models such as DeepSeek, The company's AI application products adopt the MOE (Hybrid Expert) architecture, and will continue to access different AI large models to provide functional enhancements and supplements to the products. On the basis of "smart", Tom Cat AI voice emotional companion robot pays more attention to the user's emotions and psychology, and provides users with the company of high emotional value of "high IQ and emotional intelligence".

According to Cha, industrial investment funds such as Zhuhai Hengqin Caixun Science and Technology No. 1 Venture Capital Partnership under Caixun Co., Ltd. have invested in Hangzhou DeepSeek Technology Co., Ltd. (DeepSeek). This kind of investment is a financial investment, and as an indirect shareholder, ColorSense does not directly participate in the daily operation and management of DeepSeek. Is it true?

Caixun [300634]: Caixun and its subsidiaries do not have any financial investment in Deepseek.

What are the company's layouts in the AI field? Can AI be used to reduce costs and increase efficiency?

Daily interaction[300766]: As the industry enters the era of large models, the company has access to ChatGPT, Wenxin Yiyan, ChatGLM, Tongyi Qianwen, DeepSeek and other general large models at home and abroad for the first time, and combined with its own professional capabilities in data accumulation and mining and deep insight into specific industries, it has developed industry vertical models applied in the fields of commercial marketing and public services, providing customers with multi-scenario solutions to improve quality and efficiency. At present, the company has applied the capabilities, concepts and methods of large models in the fields of digital intelligent transportation, public safety and digital marketing, and in the next step, the company will continue to explore the innovative application of new technologies such as artificial intelligence in different scenarios, and launch more professional and subdivided industry vertical models to maximize the value of data.

The

relationship between Yitong Century and DeepSeek is mainly reflected in the fact that Yitong Century participated in the financing activities of DeepSeek as an investor, and the participation of Yitong Century provided additional financial support for DeepSeek. Is it true? What is the investment amount?

Yitong Century [300310]: The information you provide is not true.

Your company participated in DeepSeek's Series A financing round in 2023 and became its strategic partner. According to reports, the company invested 20 million yuan in DeepSeek with its own funds, with a shareholding ratio of 0.6667%. What is the strategic purpose of the company's investment?

Century Tianhong [300654]: At present, the company does not have any equity investment relationship with Deepseek, nor does it have any other cooperative relationship.

At

11:22 on January 27, 2025, the official account of China Fund News published an article "Just now! Sudden detonation, multi-share one-word limit", it is said that Huajin Capital indirectly holds shares in Deepseek, and Huajin Capital participates in Deepseek's Pre-A round of financing through Huajin Lingyue Fund, is it true?

Huajin Capital [000532]: On February 6, 2025, the Company disclosed the "Announcement on Abnormal Fluctuations in Stock Trading" (Announcement No.: 2025-003) on the information disclosure media selected by the Company, in which it responded to the recent rumors related to the DeepSeek hotspot and the Company: The Company (including its holding subsidiaries) and the funds managed by the Company's subsidiaries did not participate in the financing of DeepSeek.

Baidu Al Q&A and Doubao Answer both said that Lihe has participated in DeepSeek through its investment institutions, with a shareholding ratio of 2%, is it true?

Lihe Kechuang [002243]: The company and its subordinate investment institutions do not participate in DeepSeek.

Have you had any technical exchanges or cooperation discussions with AI companies such as DeepSeek? Is there a specific plan for the field of AI+GIS (such as remote sensing analysis and data automation)?

SuperMap Software [300036]: The company plans to adapt and access DeepSeek's latest R1 series large model as the underlying model to empower the company's AI large model products, in addition, the company will fine-tune and train the company's industry model products based on the DeepSeek training program. The relevant access will accelerate the maturity of the company's AI large model products, including the ability of the company's geospatial agent products to understand user intent, make decisions, make judgments, and call tools.

The emergence of deepseek has greatly reduced the cost of AI applications, and many cloud computing companies have laid out deepseek and built it on the cloud.

Shunwang Technology[300113]: Shunwang Intelligent Computing is an AI INFRA platform, which supports comprehensive scheduling, model inference deployment, and API invocation based on the computing power of Shunwang cloud edge computing power and the computing power of the intelligent computing center. At present, the company will continue to promote the innovation and improvement of its own products and technologies according to the development trend of related technologies, and continue to provide convenient and efficient artificial intelligence solutions for developers and enterprise users.

Does your company have any deepseek related business and investment?

Jiuyuan Yinhai [002777]: Through the access to DeepSeek-R1 and the integration of its related capabilities, the company has greatly improved the quality and efficiency of the training of the company's Yinhai "Wenyu" large model. In the vertical scenarios of digital government, medical insurance, and smart city, it can provide customers with customized industry intelligent solutions such as government intelligent services, government governance innovation, government smart decision-making, medical AI knowledge base, smart medical services, and smart hospital management. The Company currently has no other business or investment partnerships with DeepSeek.

Does Xiechuang Cloud have any plans to launch Deepseek?

Xiechuang Data [300857]: The company has recently released the FCloud intelligent body training and promotion innovation platform product, and serves many enterprises such as Zhangjiang Group, the platform has been connected to deepseek and other large models, providing customers with one-stop computing power training and inference services.

Does the company's cloud computing currently have plans to access deepSeek, or access other large AI models? How will the use of DeePSeeK benefit the company's operations?

Fengyuzhu [603466]: At present, the company has access to ChatGPT, Wenxin Yiyan, Doubao AI and DeepSeek and other general large models at home and abroad, and has completed the privatization deployment in combination with its own content creative needs and data security needs, which is expected to effectively assist the company's creativity, planning, design, CG rendering and 3D content generation, interactive technology and other teams to provide customers with quality and efficiency services.

As a local company in Hangzhou, will the company give priority to cooperating with DeepSeek in large-scale model technology?

Taxyou Co., Ltd. [603171]: At present, the company's AI application framework can be compatible with domestic mainstream large language models. The company will start to connect and trial the DeepSeek open source model and API interface from October 2024. Because the DeepSeek large model has significant cost advantages in some scenarios, the model has been applied to the company's billion-dollar SaaS platform agent consulting services, digital intelligent operation and maintenance and other scenarios. At present, the company uses about 4 billion DeepSeek large model tokens, which accounts for a relatively low proportion of all large model usage of the company, so investors should pay attention to investment risks.

The

application of large models in the mining industry has always been an important R&D direction of Longsoft, and the open source of DeepSeek large models has brought our country's AI development to an unprecedented new height.

Dragonsoft Technology [688078]: "Dragonsoft Technology has successively realized the docking with DeepSeek-V2 and DeepSeek-R1 large models, and successfully completed the access of large models to core products such as Longsoft intelligent management and control platform and Longsoft Smart Map. Based on the experience and data precipitation of hundreds of pairs of projects such as intelligent management and control platform for production mines and transparent geological guarantee completed by Longsoft Technology, the company fully excavates the innovative application of large model capabilities in the daily safety production control and other businesses of mines, and continuously upgrades the intelligent core of mine management and control through a series of technological innovations such as "intelligent map viewing, intelligent data checking, and intelligent decision-making suggestions", and builds an AI application-side capability system in the mining field. At present, the company's management and control platform based on the DeepSeek large model has begun to be applied in relevant implementation projects, and will gradually complete the transformation of other large models that have implemented intelligent mining business systems, so as to provide guarantee for the digital and intelligent transformation of the industry. Since the company started the research and application of artificial spatiotemporal intelligence in 2023, the strategic layout of AI has achieved initial results. In the future, Longsoft Technology will give full play to the business, scenario, and massive data accumulation of more than 20 years in the mining industry, focus on the characteristics of spatio-temporal intelligence, carry out the R&D and application of general spatio-temporal intelligence large models (serving language processing, machine vision, industrial engineering, robotics, unmanned driving, etc.) and the core technologies of mine vertical large models, and collaborate with excellent general large model ecosystems including DeepSeek to provide the industry with full-stack AI solutions from knowledge base to decision support. Promote the intelligent reshaping of mine safety production mode. ”

On February 5, 2025, Silicon-based Flow, in which Zhipu Huazhang is a shareholder, announced that it has successfully launched the new DeepSeek model with HUAWEI CLOUD in the Ascend community. Does your company currently hold shares in Zhipu Huazhang? Is it the previous investment of 160 million yuan?

Lingyunguang [688400]: Zhipu Huazhang is an outstanding representative of a new generation of cognitive intelligence model, the company has invested in the company in a small proportion, and the two sides have maintained a good business cooperation relationship in the fields of AIGC digital content intelligence and industrial big data.

Seeing that some shareholders reported that the company's Lotus Purple Star has successfully deployed the DeepSeek R1 model, is the news true?

Lotus Holdings[600186]: The company's holding subsidiary, Lotus Zixing, has successfully deployed the DeepSeek R1 model on its own computing power and supported the invocation of the Lotus Zixing agent platform. Founded at the end of 2023, Lotus Purple Star is still in its infancy in the field of AI and computing power applications. The deployed DeepSeek R1 is an open-source large model for deep seeking, and multi-user Q&A is supported on the agent platform. The company has no cooperation with DeepSeek at the capital or equity level, and the deployment of the DeepSeek R1 large model is more common in the field of AI, which is uncertain about the company's product development and the progress of industry scenarios.

Does your company cooperate with Deepseek? What kind of business cooperation exists, and what is the market size and market prospects? Competing companies in the same business with Deepseek?

Nanwei Software [603636]: The company has carried out product integration and capability invocation based on the Deepseek open source model, which has effectively improved the AI capabilities of products in the field of digital government. The company believes that Deepseek has efficient characteristics in the use of computing power, which is conducive to the application of large models in more fields. At the same time, for government departments, it can improve the AI capabilities of government informatization at a lower cost. Especially in the case of high government requirements for data security, Deepseek is easier to deploy privately, which can help accelerate the upgrade of digital government. Compared with the current large models of large manufacturers with high dependence on Internet data, Deepseek's privatization deployment capability is its core competitive advantage.

The DeepSeek model has shocked the world, and with the reduction of the cost of large model invocation, the reliability and security of large models are more important. Has the company's digital wind tunnel project also been connected to deepseek large models, and will it deepen cooperation with large model companies in the future to provide safe and reliable AI services for the majority of users?

Yongxin Zhicheng [688244]: The company's AI large model security evaluation "digital wind tunnel" platform (hereinafter referred to as the "platform") has been connected to the DeepSeek series of large models. On February 5, the company's official account released "From the first day of the Lunar New Year to the seventh day of the Lunar New Year, Yongxin Zhicheng AI's "Digital Wind Tunnel" team tackles tough problems continuously: what level of DeepSeek is answered? How to make it clear? Can an enterprise budget of 100,000 yuan be deployed? The platform has loaded more than 400,000 test data compatible with international evaluation standards such as the European Union's "Artificial Intelligence Law", the United States' NIST "Artificial Intelligence Risk Management Framework", and WDTA AI-STR-02 "Large Language Model Security Testing Method", with more than 100 basic datasets, more than 5 million total evaluation use cases, more than 20,000 AI jailbreak and variability detection payloads simulating the red team, and more than 100W security detection data and more than 2W attack payload templates. The platform has conducted special tests on it from the dimensions of intelligence, security, matching, and consistency, and its comprehensive evaluation results are leading. The company has been actively promoting cooperation with large-scale model enterprises, please continue to pay attention to the relevant information disclosed by the company for specific progress.

How many rounds of funding has your company participated in DeepSeek?

Jinhai Hi-Tech[603311]: Neither the Company nor the participating funds directly or indirectly participate in the financing of DeepSeek.

Does the company's air-ground integrated large model research and development and intelligent body robot research and development cooperate with Unitree Technology, a strategic partner in the news report? IS DEEPSEEK TAKING ADVANTAGE OF THIS HISTORICAL OPPORTUNITY TO ACCELERATE THE ADAPTATION OF THE MODEL TO RAPIDLY PROMOTE THE COMPANY'S AI STRATEGY? At the same time, the open source model can also quickly adapt to the research and development of unmanned driving-related vertical models.

Austel Technologies[688208]: Autel Technologies has completed the full integration and localization deployment of Deepseek and applied the DeepSeek training process to accelerate the wide promotion of Autel Deepseek's air-ground integrated solution with the inspection vertical large model as the core in the fields of energy and transportation. The company's AI team has deep experience in technology accumulation and successful commercialization, and we continue to cooperate in the application of cutting-edge basic large model algorithms at home and abroad, including Chatgpt, DeepSeek, Llama, Qwen, etc., conduct targeted vertical model training according to business needs, and carry out in-depth algorithm innovation and application development, so as to build leading vertical large models such as automotive intelligent maintenance, intelligent charging and energy management, inspection and embodied intelligence.

According to news reports, Wondershare Technology is evaluating and exploring the landing direction of the DeepSeek-R1 large model on the application side. Can you tell us about the progress and the impact on your company? At such a critical moment, did your company give up the vacation and implement it as soon as possible?

Wondershare Technology [300624]: The company has completed the adaptation of the inference model DeepSeek-R1, and has implemented the integration of related capabilities in Wondershare MeowFilm, Edraw Icon, Edraw MindMaster, Wondershare PDF and other products.

OpenAI is about to launch the former Q project codename strawberry model, it is said that all the current large models advanced to the L2 level, your company previously released the Q algorithm with the National University of Singapore, claiming to greatly improve the efficiency of the existing model, can break the monopoly of Q, is the algorithm currently mature and integrated into your company's large model? Can you also tell us about your company's large-scale model development plan?

Kunlun Wanwei [300418]: Kunlun Wanwei and Nanyang Technological University in Singapore have successfully developed the Q* algorithm, which significantly improves the inference ability of existing large models. On the GSM8K dataset, Q* helped Llama-2-7b improve its accuracy to 80.8%, surpassing ChatGPT. On the MATH dataset, Q* helped DeepSeek-Math-7b achieve an accuracy rate of 55.4%, surpassing that of Gemini Ultra. On the MBPP dataset, Q* helped CodeQwen1.5-7b-Chat improve its accuracy to 77.0%, narrowing the programming gap with GPT-4. Q* can not only greatly improve the performance of small models, but also significantly reduce the demand for computing resources, bringing new possibilities for the wide application of artificial intelligence. The company's Tiangong model also continues to iterate.

The MACD golden cross signal is formed, and these stocks are on a good rally!