[Table_Page] 跟踪分析|计算机 证券研究报告 [Table_Title] 计算机行业 GenAI 系列（二十七） Token 高速增长的背后：应用突破，与算 力同享加速发展机会 [Table_Summary] 核心观点： | [Table_Grade] 行业评级 | 买入 | | --- | --- | | 前次评级 | 买入 | | 报告日期 | 2026-03-01 | [Table_PicQuote] 相对市场表现 -20% -11% -2% 6% 15% 24% 03/25 05/25 07/25 10/25 12/25 02/26 计算机 沪深300 | [分析师： Table_Author]刘雪峰 | | --- | | SAC 执证号：S0260514030002 | | SFC CE No. BNX004 | | 021-38003675 | | gfliuxuefeng@gf.com.cn | | 分析师： 周源 | | SAC 执证号：S0260523040001 | | 0755-23948351 | | shzhouyuan@gf.com.cn | | 请注意，周源并非香港 ...

这是 时隔21个月 以来的最大版本的更新。 金磊 发自 凹非寺 量子位 | 公众号 QbitAI 在 Seedance 2.0 和 Seedream 5.0 Lite ，一波接一波爆火之后，豆包把完全体拿出来了—— 豆包大模型2.0 。 像Seedance 2.0已经成为全民玩转的AI，我们也试着做了一个视频： 短短5秒钟，效果确实是足够逼真。 也难怪老外也开始研究怎么注册中国手机号来体验了…… 再如 Seedream 5.0 Lite ，首次支持联网检索，生成的图片也达到了商业化的水平： 而就在今天，在视觉模型火爆之后，豆包终于把那个最核心的大脑拿出来了—— 豆包大模型2.0 。 整体来看，这次豆包大模型2.0在多模态理解、企业级Agent、推理和代码能力上都有了不少的提升： 更直观的提升，体现在榜单测评中。 例如在MathVista、MathVision、MathKangaroo、MathCanvas等数学推理基准上达到业界最优水平。同时，在 LogicVista、VisuLogic 等视觉解谜与逻辑推理基准上，Seed2.0 Pro得分较Seed1.8显著提升。 更强多模态理解：在多模态感知、高精度文字 ...

Investment Rating - The report maintains a positive outlook on the computer industry, particularly highlighting the advancements in AI technologies and their implications for enterprise applications [5][6]. Core Insights - OpenAI's release of GPT 5.2 marks a significant improvement in reasoning and professional knowledge capabilities, achieving human expert-level performance in various assessments [7][8][11]. - The Chinese civil aviation information network is positioned as a leader in aviation IT, showing signs of recovery and growth, with a strong market presence [28][39]. - Key company updates include the acquisition activities of Gaon Electronics and the launch of TicNote Pods by Out the Door, indicating strategic expansions in their respective markets [47][51]. Summary by Sections OpenAI GPT 5.2 Release - The GPT 5.2 model has shown substantial improvements in various benchmarks, achieving 70.9% in knowledge work tasks and 86.2% in abstract reasoning tasks, outperforming previous models [8][9][11]. - The competitive landscape among major AI players like OpenAI, Google, and Anthropic is shifting from a focus on parameter size to differentiation in capabilities [7][27]. - The enhancements in reasoning and professional task execution are expected to benefit enterprise-level applications significantly [21][27]. Chinese Civil Aviation Information Network - The company is the largest provider of aviation IT services in China, deeply integrated into the information systems of the civil aviation industry [28][30]. - The recovery in the aviation sector is reflected in the increasing number of flight bookings, with a projected growth in passenger numbers and revenue [39][41]. - The company is expanding its services into the online travel agency (OTA) market, leveraging its existing data infrastructure and customer base [42][46]. Company Updates - Gaon Electronics is in the process of acquiring 100% of Ruicheng Chip Micro and 45.64% of Nengneng Micro, with a total transaction value of 21.74 billion RMB [47][49]. - Out the Door has launched TicNote Pods, a standalone AI recording device, marking a strategic shift towards independent AI hardware solutions [51][52].

Investment Rating - The report maintains a positive outlook on the computer industry, particularly highlighting the advancements in AI technology and its implications for enterprise applications [5][6]. Core Insights - The release of OpenAI's GPT 5.2 marks a significant improvement in reasoning and professional knowledge capabilities, achieving human expert-level performance in various assessments [7][8]. - The Chinese civil aviation information network is positioned as a leader in aviation IT, showing signs of recovery and growth, with a strong market presence [28][39]. - Key company updates include the acquisition activities of Gaon Electronics and the launch of TicNote Pods by Out the Door, indicating strategic expansions in their respective markets [47][51]. Summary by Sections OpenAI GPT 5.2 Release - OpenAI's GPT 5.2 was launched on December 11, 2025, showcasing substantial improvements in reasoning and professional knowledge tasks, achieving scores of 70.9% in GDPval and 86.2% in ARC-AGI-1 [7][8]. - The competition among large models is shifting from a focus on parameter size to niche differentiation, with OpenAI, Google, and Anthropic each excelling in different areas [5][27]. - The enhanced capabilities of GPT 5.2 are expected to benefit enterprise-level applications, particularly in complex reasoning and planning tasks [21][27]. Chinese Civil Aviation Information Network - The Chinese civil aviation information network is the largest provider of aviation IT services in China, deeply integrated into the industry's information system development [28][30]. - The company has a significant market share, with approximately 95% in the domestic market and around 28% globally, ranking among the top three GDS providers [35][36]. - The recovery of the aviation industry post-pandemic is reflected in the increasing number of flight bookings, with a projected growth in passenger numbers and IT investment in the sector [39][41]. Company Updates - Gaon Electronics announced the acquisition of 100% of Ruicheng Chip Micro and 45.64% of Nengneng Micro, with a total transaction value of 2.174 billion yuan, aimed at enhancing its competitive position in the semiconductor industry [47][49]. - Out the Door launched TicNote Pods, a 4G AI recording headset, marking a strategic shift towards independent AI hardware, which is expected to enhance user experience and operational efficiency [51][55].

Core Insights - The value of enterprise-level AI agents depends on having the right model, good data, and appropriate scenarios [2][3] - The complexity of real business processes can significantly reduce the success rate of AI agents, highlighting the gap between demo performance and actual operational effectiveness [2][3] Data and Model Integration - A significant portion of enterprise data, approximately 10%, is structured and contains core business logic, which is crucial for AI understanding [5][4] - The focus should not be on the largest models but rather on the most suitable ones for specific scenarios, emphasizing the importance of good data [3][5] Unique Advantages of the Alibaba Ecosystem - The background of Lingyang as part of Alibaba's data middle platform provides a unique perspective on the value of data in enterprise AI [6] - The integration of Alibaba's ecosystem allows Lingyang to leverage scenario-based data capabilities for its AI solutions [6][8] Product Design and Functionality - Lingyang's AgentOne allows enterprises to manage AI agents similarly to managing employees, offering options to recruit existing agents or train custom ones [9][10] - The product architecture consists of model services, training capabilities, and rapid agent deployment, tailored to specific business needs [10] Targeted Scenarios for AI Implementation - Lingyang prioritizes addressing the most pressing pain points for enterprises, focusing on marketing, analysis, and operational scenarios [12][14][16] - The marketing agent encompasses various functions, including creative inspiration and media placement, to optimize advertising budgets [14] Conclusion - Lingyang aims to create a comprehensive closed-loop system for enterprise data, transitioning from serving human needs to serving AI requirements, thereby enhancing the understanding and application of AI in business contexts [5][18]

Core Insights - The enterprise-level Agent market is still in a cautious phase, with many companies hesitant to adopt new technologies without clear value propositions [1][3] - A significant turning point is emerging, as companies like Qianxun Position have successfully launched AI "employees" that demonstrate tangible benefits [1][2] - The challenges of implementing enterprise-level Agents include the need for deep integration into business processes, concerns over data quality, and the importance of security and stability [3][6][9] Group 1: Challenges in Enterprise-Level Agents - Transitioning enterprises from passive acceptance to active adoption of Agents is more difficult than anticipated [3] - Enterprises prioritize value, cost, and security when considering new technologies, leading to hesitance in adopting Agents [6] - The complexity of business processes and the non-standardized nature of many enterprise scenarios complicate the implementation of Agents [8][9] Group 2: Qianxun Position's Approach - Qianxun Position has launched its first digital employee, achieving an 80% accuracy rate and a 50% completion rate in customer service queries [1] - The company plans to incubate over eight AI employees by 2025, indicating a commitment to improving the usability of digital employees [1] - Qianxun's approach emphasizes the importance of continuous learning and efficiency in customer service roles [1][2] Group 3: Insights from Lingyang - Lingyang, a subsidiary of Alibaba Cloud, is focusing on deep understanding of specific business scenarios to enhance the effectiveness of Agents [6][8] - The company has released multiple enterprise-level Agents targeting high-demand areas such as customer service and data analysis [6][7] - Lingyang's strategy involves leveraging its extensive experience and understanding of enterprise needs to create practical solutions [10][15] Group 4: Data and Integration - Effective use of Agents requires high-quality, structured data, which many enterprises struggle to manage [9][13] - Lingyang's AgentOne platform aims to streamline the development and deployment of Agents, reducing the time to market [9][20] - The integration of Agents with existing business systems is crucial for maximizing their value and effectiveness [9][20] Group 5: Future of Enterprise Agents - The evolution of enterprise-level Agents is expected to progress through three stages: basic understanding, execution capability, and self-evolution [23][25] - Companies that successfully integrate AI into their operations may emerge as "super companies," fundamentally transforming their industries [25][26] - The collaboration between humans and AI is anticipated to redefine competitive dynamics in the future [25][26]

Core Viewpoint - The focus of the model industry is shifting towards inference, with a predicted inflection point in 2025 when inference demand will surpass training demand [1] Group 1 - Tencent's MAU (Monthly Active Users) for its IMA has increased 80 times in six months [1] - The "Yuanbao" service has answered high school entrance exam questions 150 million times [1] - By 2028, it is expected that 33% of enterprise software will include Agentic AI [1] Group 2 - The domestic enterprise-level Agent (intelligent agent) market is projected to exceed $27 billion [1]

Core Insights - Tencent's IMA (Intelligent Model Assistant) has seen a significant increase in monthly active users, growing 80 times within six months [1] - The focus of the model industry is shifting towards inference, with expectations that by 2025, the demand for inference will surpass that for training [1] - The IMA has answered high school entrance exam questions 150 million times, indicating its widespread usage and utility [1] - By 2028, it is projected that 33% of enterprise software will incorporate Agentic AI, with the domestic enterprise-level Agent market expected to exceed $27 billion [1]

Group 1 - The core idea of the article is how McDonald's leverages AI and digital transformation to enhance operational efficiency and customer experience in the competitive fast-food industry [1][3][5] - McDonald's has integrated AI across various aspects of its business, including customer ordering, inventory management, and supply chain optimization, demonstrating a comprehensive approach to digitalization [2][4][6] - The company has focused on a customer-centric digital strategy since 2020, aiming to support its vast operations, which include over 7,000 stores and more than 300 million members [7][9][10] Group 2 - The introduction of the RGM BOSS system in 2022 has streamlined operations for over 200,000 restaurant employees, allowing them to save time on administrative tasks and focus more on customer service [11][12] - McDonald's IT team employs an Agile development model, enabling rapid deployment of new features and continuous improvement in business operations [13] - The event highlighted the importance of organizational adaptability and the need for businesses to identify high-value scenarios for AI implementation, such as customer service and operational processes [34][36][37] Group 3 - The article discusses the emergence of enterprise-level AI agents, which are designed to enhance business processes and improve efficiency across various functions [14][22][23] - Key insights from the event included the necessity for businesses to integrate AI into their workflows and the importance of data quality and process standardization for successful AI deployment [30][34][36] - The consensus among industry leaders was that the real challenge lies not in the technology itself but in the organization's ability to effectively utilize AI solutions [35][36][38]

Core Insights - The essence of enterprise-level AI is the role it plays in business processes, with a growing consensus that AI should be integrated into specific business functions [2][6][20] - BetterYeah is emerging as a leader in the enterprise Agent space, having recently secured significant funding and demonstrating a robust platform that focuses on process management and role definition rather than just technical specifications [5][12][15] Group 1: Market Trends and Challenges - The enterprise AI Agent market in China is rapidly developing, with projections estimating a market size of over $27 billion by 2028 [6] - Despite the enthusiasm for deploying AI Agents, the success rate for these projects is below 30%, highlighting significant challenges in integration and effectiveness [7][8] - The failure of many Agent deployments is attributed to a lack of understanding of enterprise-specific needs and the complexity of business processes [9][10] Group 2: BetterYeah's Unique Position - BetterYeah's platform is designed to address the specific needs of enterprises, offering features that allow for quick setup and seamless integration with existing business systems [15][17] - The platform has seen a 400-fold increase in AI task calls since its launch, serving over 100,000 enterprise teams, indicating strong market acceptance [12][18] - BetterYeah's approach includes a focus on real business roles and processes, differentiating it from other platforms that primarily target consumer applications [10][18] Group 3: Future of AI in Enterprises - The evolution of AI from tools to digital employees and teams is reshaping productivity in organizations, with BetterYeah at the forefront of this transformation [20][23] - Companies like Belle Group have successfully implemented over 800 AI applications using BetterYeah's platform, demonstrating the practical impact of enterprise-level AI [22] - The shift towards a collaborative model of AI, where multiple Agents work together, is expected to enhance operational efficiency and redefine roles within organizations [21][24]