If you were to name the most sensational event in the field of artificial intelligence this year, most people's answer would be the emergence of DeepSeek. In early February of this year, DeepSeek reached the top of the app download rankings in 140 countries and regions, fully demonstrating its technological confidence through its open source strategy. Open source big models refer to large-scale pre trained models developed and publicly released by research institutions or companies. Their source code, parameter weights, and even training data (or data generation methods) are open and transparent, and anyone can access, use, modify, and distribute them. DeepSeek is a completely open-source model that opens up various aspects including data, code, weights, inference chain operation ideas, and engineering construction methods, making more people willing to participate Huang Wenhong, Deputy Director of the Software Industry Research Office at CCID Research Institute, explained that it is like building a house. True open source not only discloses drawings, materials, and building structures, but also clearly tells you what to build at each step. With this information, you can return the original house 1:1. It can be seen that DeepSeek has an unprecedented level of openness, which is also one of the reasons why it has had a huge impact since its release. The biggest significance of DeepSeek for the development of China's artificial intelligence industry lies in that it has built a development ecology based on the large model of the independent research and development base. The adaptation of software and hardware, application promotion and even product promotion are all jointly completed by global manufacturers and developers, greatly reducing the cost of ecological construction. For example, global technology giants such as Microsoft, Nvidia, Amazon, Intel, and AMD have successively announced the launch of DeepSeek's open-source model inference service. Domestic manufacturers such as Tencent Cloud and Alibaba Cloud support one click deployment and calling of DeepSeek. "It's just like Android operating system is open source software. Because open source has good adaptability, mobile phones, chips, smart homes and other manufacturers have access to it, making Android a universal technology base that can keep pace with Apple IOS system in the mobile Internet era." Huang Wenhong told reporters that Internet companies, car companies and three major operators have access to DeepSeek, and many specific applications will be deployed on the DeepSeek base model later, which will rapidly increase the number of users and market share. In addition to DeepSeek, there are many open-source models in China that have attracted widespread attention in the industry. Just before the release of DeepSeeker R1, Shanghai AI startup MiniMax released its open-source model MiniMax-01, which for the first time adopted a linear attention mechanism and achieved a technological breakthrough; Alibaba's latest open-source Tongyi Qianwen QwQ-32B inference model has performed well in multiple authoritative evaluations of mathematics, code, and general abilities, ranking first on the Hugging Face trend list of the world's largest AI open-source community and becoming one of the most popular open-source models currently available. All of these demonstrate the driving role of open source in the technology ecosystem, attracting support from all parties in the industry and creating a good atmosphere for technology sharing Huang Wenhong believes that the field of information technology follows the law of constant strength among the strong. China has a first mover advantage in the open source model and must further strengthen its technological "moat". However, currently there are not many talents in various industries who truly understand open source models. Downstream enterprises based on open source models, as well as R&D personnel engaged in engineering optimization and model tuning, are relatively few. Universities should cultivate more relevant talents to jointly promote "innovation sharing re innovation" and help the industrial ecosystem become more complete. Zhu Xunyao, senior director of Alibaba Cloud, believes that the open source concept has not yet reached a broad consensus in the industry, but the success of DeepSeek and Tongyi Qianwen will gradually make everyone realize that the open source model will become the most powerful engine to promote the development of AI in China. Next, it is recommended to embrace open source with a more proactive attitude, from the national level to the local level and even to enterprises. At the same time, we should accelerate innovation in areas such as deploying intelligent computing power, building high-quality datasets, and using cloud computing to keep up with the world's advanced level. Since the release of DeepSeek, various industries have been exploring the integration of it into their own business scenarios. The open-source model, with its technological advantages of low cost, high performance, and high openness, has accelerated the popularity of artificial intelligence in the industry, "said Huang Wenhong. In February of this year, the Hang Seng Electronics big model application was fully integrated into DeepSeek, achieving good results in business scenarios such as financial investment research, compliance, operations, and investment banking. For example, in the investment banking business, utilizing DeepSeeker R1's understanding ability can automatically parse complex documents such as prospectuses and due diligence reports, achieving instant response to financial data verification and compliance risk alerts. The Tongyi Qianwen open source model Qwen series, with its multimodal and full-size technical capabilities, as well as a good ecosystem gathered by a large number of developers and small and medium-sized enterprises, accelerates the empowerment of various industries. As of now, Alibaba has open sourced over 200 models, including text generation models, visual understanding/generation models, speech understanding/generation models, text and video models, and other multimodal models, covering various sizes ranging from 0.5B to 110B parameters. Last April, the Artificial Intelligence Working Group of the National Astronomical Observatory of the Chinese Academy of Sciences released a new generation of astronomical model "Star Talk 3.0" based on Qwen. At present, it has successfully connected to the Mini "Sitian" telescope array of the Xinglong Observatory of the National Astronomical Observatory, which can autonomously control the telescope for observation, analyze observation results, and intelligently provide next observation suggestions. This is the first application of large models in the field of astronomical observation. From predicting protein structures to synthesizing targeted drugs, to discovering new types of viruses, the combination of big models and scientific research has brought many breakthrough achievements, "said Zhu Xunyao. Whether it's DeepSeek or Tongyi Qianwen, China's open-source models are improving the breadth of artificial intelligence applications in the industry with their relaxed development licenses and low-cost training methods. The business model still needs to clarify the closed source model represented by ChatGPT, which is called the "token economy", that is, by providing API services to users, pricing based on token usage, and then earning profits. So, how do open source models profit? Regarding this, Huang Wenhong shared several cases with reporters. The Llama big model launched by Meta can attract more enterprises and developers to join its ecosystem through open source, creating opportunities for future advertising revenue. Open source, closed source, and parallel model products have also emerged in the market. Specifically, first open up relatively basic capabilities and cultivate user usage habits, while higher performance models require payment for use. Some open-source models will be bundled with cloud services for sale, which means the models are free and only charge for computing power. This model is like having to equip an Apple phone if you want to use the IOS system. Another more similar case is that Google attracts users through the Android system and charges fees by selling value-added services such as Google Mail and Google Maps, "Huang Wenhong added. In Zhu Xunyao's view, many companies that create open source models have a strong technical idealism, and their original intention may not be entirely towards commercialization. The Tongyi Qianwen big model has over 100000 derivative models and hundreds of millions of downloads worldwide, all of which are provided to users for free. However, because of open source, a large number of developers are attracted, and Alibaba Cloud's model services and supporting computing services are favored by more developers. Developers and manufacturers form a virtuous cycle of 'open source application feedback'. The development of open source models is still in its early stages, and the industry is still exploring how to form a healthy and mature business model. Enterprises definitely want to make profits by creating open source models. They need to explore a positive cycle development path, find a balance between technology inclusiveness and commercial monetization, and ensure that all participants in the industry chain can benefit, ensuring the continuous and stable operation of the open source model, "suggested Huang Wenhong. (New Society)