Guolian Minsheng Securities: AI Agent Manus is born, model equality promotes the development of AI industry chain.

Guolian Minsheng Securities released a research report, stating that recently, the domestic Monica team has released the AI Agent product Manus. Manus adopts multiple model dynamic calls and does not independently develop large models, instead using different large models in combination. The company emphasizes integrating and optimizing resources to meet customer needs as a product strategy. The emergence of Manus signifies a rapid development in the AI application market following model equality. Since January 2025, the efficient reasoning model DeepSeek R1 has been open-sourced, starting the era of model equality. Companies planning to implement AI empowerment can access or deploy locally through API interfaces, facilitating model empowerment advancement and potentially accelerating the development of the application market. Key points from Guolian Minsheng Securities: Industry events On March 6, 2025, the domestic Monica team released the AI Agent product Manus, which can automatically execute programs after receiving commands. Manus not only provides suggestions or answers but can also deliver complete task results and display the entire reasoning process. This AI Agent combines Claude's Computer use with OpenAI's Deep Research and can generate code directly, becoming a coding Agent. Manus surpassed OpenAI's Deep Research in the GAIA (General AI Assistant) benchmark evaluation, ranking first in GAIA scoring. Manus embodies the functionality of an Agent through a long chain of thoughts and tool calls. Manus includes three cases in the promotional video: the first involves uploading a resume to Manus for screening, where Manus reads the resume, provides ranking suggestions, and generates a table upon request, demonstrating the entire process to the user; the second case is property search, where Manus can filter properties based on user requirements; and the third case is stock analysis, where Manus can open Python for analysis and generate web files. In addition to showcasing video functions, Manus can handle agent tasks in fields such as travel planning, educational content innovation, insurance policy comparison, supplier procurement, financial report analysis, startup company list organization, and online store operation analysis. DeepSeek R1's open-source approach empowers application companies with model equality. Manus uses multiple model dynamic calls and does not independently develop large models, opting to use a combination of different large models. The company stresses integrating and optimizing resources to meet customer needs as a product strategy. The emergence of Manus signifies a rapid development in the AI application market following model equality. Since January 2025, the efficient reasoning model DeepSeek R1 has been open-sourced, starting the era of model equality. Companies planning to implement AI empowerment can access or deploy locally through API interfaces, facilitating model empowerment advancement and potentially accelerating the development of the application market. Model equality empowers the AI industry chain, leading to rapid growth in computing power, public cloud, and integrated machine markets. The open-source release of efficient and low-cost reasoning models has driven the overall development of the AI industry chain. With model equality, data, computing power, models, and application segments of the industry chain are expected to experience rapid development, leading to investment opportunities in four areas. (1) Opportunity in computing infrastructure construction: Improved large-model capabilities or changes in computing power demand are driven by the open-source models, accelerating the development of the domestic GPU ecosystem. Leading companies to consider include Dawning Information Industry (603019.SH), Unisplendour Corporation (000938.SZ), and Inspur Electronic Information Industry (000977.SZ). (2) Opportunity in AI software development on the edge: Edge devices will be crucial for human-machine interaction, potentially driving opportunities in software development for edge AI applications. Companies to watch include Thunder Software Technology (300496.SZ) and related companies. (3) Opportunity in productivity tools innovation: Large models are expected to reduce professional barriers and repetitive labor for productivity tools, with potential benefits for companies like Beijing Kingsoft Office Software, Inc (688111.SH), Yonyou Network Technology (600588.SH), and Weaver Network Technology (603039.SH). (4) Opportunity in industry informationization innovation: Large models are expected to first be implemented in industries such as finance, law, education, healthcare, and automotive. Industry informationization vendors will play a crucial role in connecting basic model manufacturers and industry customers. Recommended industry informationization head enterprises include Hundsun Technologies Inc. (600570.SH), Yusys Technologies (300674.SZ), Zhejiang Supcon Technology Co., Ltd. (688777.SH), Winning Health Technology Group (300253.SZ). Risk warning: Commercialization progress may fall short of expectations.Risk: The application of AI large models is still in the early exploration stage, mature profit-making business models have not yet been formed, and there may be risks in the commercialization process that are lower than expected.Data Security Risks: The operation of AI large models requires a large amount of data, which may contain personal information of users, companies' trade secrets, and even sensitive government data, posing risks of leaking personal information and sensitive data. Intensified Industry Competition Risks: Participants in the field of AI large models include emerging startups, internet giants, vertical IT companies, etc., creating risks of intensified industry competition.