Riding the wave of breakthroughs in domestic technology and large models, China's banking industry is launching a self-reform wave driven by AI large models. In this transformation, the DeepSeek large model acts like a "fintech scalpel," incorporated into the technological foundation of major banks such as Industrial and Commercial Bank of China (ICBC), Postal Savings Bank of China (PSBC), and Shanghai Pudong Development Bank (SPDB), and has also become a part of internet banks like Xinyong Bank and Zhongbang Bank. SPDB's fully domestic deployment to Guilin Bank's "Smart Cube" for cost reduction and efficiency enhancement, from the complex reasoning of ICBC's financial report analysis assistant to the second-level decision-making of Qingnong Commercial Bank's "Smart Qimi" middle platform, this efficiency revolution driven by large models may have the opportunity to reconstruct the service paradigm of the banking industry in the future. ICBC: Unified Entry, Rich Large Model Matrix With the help of DeepSeek, the state-owned major bank has further enriched and enhanced its self-controllable large model platform. ICBC stated that it has previously taken the lead in establishing a fully self-controllable trillion-level financial large model technology system, creating the "ICBC Smart Surge" large model brand, and building a "commercial + open-source" large model matrix for the financial industry, leading the industry's innovative applications. In the first half of 2024, ICBC has begun to deploy and pilot the DeepSeek series of open-source large models internally. The latest open-source large model introduced from DeepSeek possesses industry-leading complex reasoning capabilities, further enriching ICBC's large model matrix, providing a unified entry through ICBC Smart Surge, and offering more efficient, convenient, and secure AI productivity tools for all employees, realizing the widespread application of AI technology throughout the bank. During the introduction of the latest open-source large model from DeepSeek, ICBC significantly improved model deployment efficiency through a series of technological innovations, reducing reasoning costs and increasing throughput. By piloting in key areas first, ICBC achieved application breakthroughs, fully leveraging the advantages of DeepSeek's open-source large model in strong reasoning and complex data processing, building over 10 typical scenarios such as financial report analysis assistants and AI wealth managers, promoting the intelligent upgrade of business processes and effectively enhancing work quality and efficiency. PSBC: Local Deployment and Integration at First Opportunity Postal Savings Bank of China (PSBC) relied on its own large model "Youzhi" to locally deploy and integrate the DeepSeek-V3 model and the lightweight DeepSeek-R1 reasoning model at the first opportunity. The "Youzhi" large model will be further enhanced in aspects such as complex multimodal and multi-task processing, computing power savings, and efficiency improvements by introducing and applying DeepSeek capabilities. PSBC first applied the DeepSeek large model to the "Xiao You Assistant," adding logical reasoning functions to enhance precise service efficiency; through deep analysis and other functions, it accurately identifies user needs and provides personalized and scenario-based service solutions; leveraging efficient reasoning performance, it accelerates response speed and task processing efficiency, providing users with a smoother interaction experience PSBC will further explore more specialized service applications in financial scenarios based on the "Youzhi" large model, leveraging DeepSeek's technical capabilities. In the field of remote banking services, it will enhance the capabilities of mobile banking companion digital employees through multi-step reasoning optimization, optimize seat assistants and intelligent practice partners, and improve customer service professionalism and work efficiency; in the field of corporate finance, it will explore scenarios in the construction industry supply chain, completing complex business reasoning processes, product recommendations, product portfolio recommendations, equity analysis, and financial analysis; in the field of risk prevention and control, it will efficiently analyze and automatically generate case analysis reports, improve the efficiency and accuracy of suspicious point identification, and enhance proactive defense capabilities against fraud; in the field of branch financial services, it will explore innovative applications on the AI edge, expand digital teller service scenarios, promote smart operations in branches, and enhance customer experience. Shanghai Pudong Development Bank: First to Achieve Full-Stack Domestic Computing Power Platform + DeepSeek Large Model Shanghai Pudong Development Bank has deployed the DeepSeek-R1 671B trillion-level large model on Ascend servers, becoming the first in the industry to achieve financial applications with a full-stack domestic computing power platform + DeepSeek large model. Specifically, Shanghai Pudong Development Bank quickly completed the deployment of the DeepSeek-R1 671B large model based on the Ascend Atlas 800T A2 computing power cluster, integrating the "five horizontal and six vertical" enterprise-level knowledge base, embedding digital employee assistant applications, and empowering multiple application scenarios such as intelligent Q&A, indicator Q&A, financial analysis, and report writing. This better empowers employees internally and provides better service to customers externally, comprehensively enhancing digital financial service capabilities. Jiangsu Bank: Actively Using Large Models to Reshape Financial Service Models Jiangsu Bank relies on the "Smart Xiao Su" large language model service platform, successfully localizing and deploying fine-tuned DeepSeek-VL2 multimodal models and lightweight DeepSeek-R1 inference models, which are used in intelligent contract quality inspection and automated valuation reconciliation scenarios, reshaping financial service models through the mining and analysis of massive financial data, achieving breakthroughs in both financial semantic understanding accuracy and business efficiency, injecting strong momentum into business development. It is reported that Jiangsu Bank was the first to research and develop the industry-level large language model service platform "Smart Xiao Su" in 2023. By introducing the DeepSeek large language model, "Smart Xiao Su" has further enhanced its capabilities in handling complex multimodal and multi-task scenarios, computing power savings, and efficiency. The DeepSeek-VL2 multimodal model can simultaneously process various data types such as text, images, and voice, saving computing power costs compared to single-domain model deployments, providing a technical foundation for further solving complex multimodal scenario problems in the financial field ( such as invoice recognition, contract analysis, etc.); the DeepSeek-R1 model has significant advantages in model scale and performance, providing better solutions for handling complex tasks (such as risk assessment, investment analysis) and generating high-quality text (such as report writing, compliance review) Internet Banks: Actively Upgrading Intelligent Customer Service Internet banks, which inherently possess technological advantages, are clearly among those actively embracing change. CITIC Baixin Bank has been vigorously promoting the application of large model technology for intelligent customer service assistants and intelligent risk control in various scenarios since the beginning of 2023. With the strong support of the DeepSeek series models, CITIC Baixin Bank's intelligent code assistant capabilities have been newly upgraded, enhancing research and development efficiency and quality. Additionally, based on user feedback, the algorithm team continuously fine-tunes the model, training it with questions and answers from actual business scenarios within the bank, making the model more aligned with CITIC Baixin Bank's research and development needs in terms of language style and business logic. As of now, the average weekly number of inquiries has exceeded one thousand, with a response accuracy rate of 91%, making it an indispensable assistant for researchers. Xinwang Bank is actively embracing large models like DeepSeek to create an intelligent service platform based on large models called "Shijuan." By deeply integrating multi-modal data processing capabilities, it addresses the challenges of personalized credit for new citizens and the complex processes of mortgage services like car and house pledges, improving service quality and efficiency. Currently, Xinwang Bank is rapidly advancing its exploration and practice of large models through efforts in strategic development, application scenarios, model construction, and computing power establishment. In multiple fields such as intelligent customer service, post-loan management, intelligent marketing, intelligent research and development, and risk control, Xinwang Bank has built a series of intelligent assistants relying on large model technologies like DeepSeek, deeply applying them to various efficiency bottlenecks. Zhongbang Bank completed the localized deployment of DeepSeek on February 6, 2025, with the first phase starting applications in scenarios such as ChatOPS, intelligent recruitment, and inquiry assistants. ChatOPS is an intelligent IT service assistant created by Zhongbang Bank based on open-source large models, deeply integrating operational knowledge bases. Currently, ChatOPS has been upgraded to the DeepSeek R1 model and integrated into WeChat Work, allowing users to access efficient services with one click. It is reported that the upgrade has achieved a leap in efficiency, with problem-solving efficiency improved by 50%, response speed shortened to seconds, and labor costs reduced by 30%; moreover, it is precise and reliable, with an answer accuracy rate exceeding 85%, and the fault handling cycle compressed by 60%, ensuring business continuity. Rural Commercial Banks: First Batch of Typical Scenarios Recently, Qingdao Rural Commercial Bank (hereinafter referred to as Qingnong Bank) has been deeply promoting the innovative application of large model technology in the financial sector, with the localized deployment of the enterprise-level AI model service platform "Smart Qimi" based on the DeepSeek large model. It is applied in scenarios such as digital humans in branch halls and training material text verification, reshaping external service and internal management models, and further promoting the intelligent upgrade of digital finance development across the bank. By introducing the DeepSeek-R1 large language model and combining it with existing large model technology capabilities, "Smart Qimi" has achieved significant breakthroughs in model lightweight and efficient inference, providing better solutions for complex and high-frequency business scenarios Based on the model characteristics of DeepSeek, Qingdao Rural Commercial Bank's "Smart Qimi" has launched the first batch of six scenarios including intelligent Q&A, report writing, and article verification. Beijing Rural Commercial Bank continues to delve into the research and application of large model technology, accumulating rich technical reserves and application experience, and has conducted research on the topic of "Application of Open Source Large Models." Through active exploration, AI technology has been integrated into business and internal management chains, covering multiple application scenarios such as intelligent interactive Q&A, intelligent extraction of work order information, intelligent question bank construction, code-assisted development, and automated test case generation, significantly enhancing business efficiency and innovation capability. Risk Warning and Disclaimer The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at one's own risk