国信证券：模型架构继续演化 多模态+长文本为Agent爆发提供基础

Group 1 - The core viewpoint of the report emphasizes the evolution of model architecture, with multimodal and long-text capabilities laying the foundation for the explosion of Agents in the AI sector [1] - The report highlights that the commercial paths of large model vendors are diverging, with a significant increase in demand for reasoning expected by 2026, which will reshape the SaaS market landscape [1] - The analysis of the stock price trends of major US tech giants over the past three years shows a continuous progression of the AI narrative, with OpenAI leading the acceleration in 2023 and Microsoft benefiting from its exclusive partnership [1] Group 2 - The report discusses the ongoing evolution of model architecture, noting that the next generation of models must address two core pain points: the computational and memory consumption bottlenecks during the training phase, and the limited memory capacity during inference [2] - It is projected that the Scaling Law will continue to be relevant, with advancements in pre-training, post-training, and reasoning scenarios, while reinforcement learning is expected to become a key breakthrough area [2] - The report indicates that the gap between Chinese and US models is currently around 3-6 months, with computational power and algorithms being critical for catching up [2] Group 3 - The report identifies that no clear winner has emerged in the general large model capabilities, with different vendors pursuing distinct commercialization paths [3] - OpenAI is noted for its strong consumer base of 800 million users, while Gemini is recognized as the current state-of-the-art (SOTA) benchmark due to its commitment to a native multimodal approach [3] - Anthropic is highlighted for its focus on the B2B market, achieving a valuation of $350 billion, while Grok is expected to leverage Tesla's unique data advantages for its next-generation models [3] Group 4 - The report anticipates that the demand for AI applications will continue to grow, with the software development landscape being reshaped by large models, which are expected to open up new ceilings for software demand [4] - It cites IDC data projecting the global SaaS market to reach nearly $1 trillion by 2029, a significant increase from $580 billion in 2025, although it notes that the competitive landscape among players will be reshuffled [4] - The report observes that large model vendors are beginning to collaborate with B2B software service providers to develop more industry-specific demands [4] Group 5 - The report predicts an explosion in demand for reasoning capabilities by 2026, with AI programming, AI Agents, and AI content creation being the primary application areas driving growth [5] - It highlights the rapid growth of several AI applications, including AI programming software Cursor, which has reached an ARR of $1 billion, and AI agent Manus, which achieved $100 million in ARR within eight months [5] - The report suggests that as model capabilities mature, there will be noticeable growth in AI applications in consumer devices and enterprise distribution channels [5]