从智谱与 MiniMax 看大模型商业化路径

Investment Rating - The report maintains a "Positive" investment rating for the industry [10] Core Insights - The Chinese large model industry is transitioning from a "hundred model war" to a tier of leading players by 2025, with Zhipu AI and MiniMax representing different commercialization paths. Both companies are set to IPO in Hong Kong on January 8 and 9, 2026, marking them as the first publicly listed large model companies globally, which serves as a significant industry benchmark [4][19] - The development paradigm of large models is moving into a new phase, shifting from early parameter competition and task performance comparisons to a core focus on a "scene-data-model" positive cycle. The ability to deeply root in real scenarios, provide smoother product experiences, and foster tighter ecological collaboration will determine competitive advantages in the next stage [4][9] Summary by Sections Historical Review: Genetic Differences Determine Commercialization Paths - Zhipu AI, founded in 2019, is one of the earliest companies in China to develop large language models (LLMs) and the first to release a self-developed large model. Its development is characterized by a strong platform thinking and aims to maximize the utilization of its technical assets through an ecosystem of developers [7][20] - MiniMax, established in 2021, has a strong "commercial product" gene, with a team background in gaming and content communities, giving it advantages in user experience (UX), retention rates, and multimodal interactions [7][42] Current Status: Commercialization Progress of Zhipu and MiniMax - Zhipu focuses on general large model capabilities, emphasizing deep integration with enterprise business processes, positioning itself as an "AI infrastructure provider." In contrast, MiniMax aims to productize model capabilities and build a user ecosystem around native AI applications, positioning itself as an "AI native content and interaction platform" [8] - Zhipu's revenue growth relies on deep engagement with B-end clients and long-term partnerships, while MiniMax's commercialization depends on user scale and activity on the C-end, utilizing subscription models and value-added services for monetization [8] Future: Moving Towards a "Scene-Data-Model" Positive Cycle - The key to future development is constructing a closed loop driven by real scene demands, generating high-quality data that continuously feeds back to optimize model capabilities, ultimately serving the scenes more accurately and creating a dual value loop of commercial and technical benefits [9] - Both Zhipu and MiniMax have entered the "scene-data-model" positive cycle, with Zhipu focusing on B-end clients and edge entry points, while MiniMax concentrates on creative content generation and enterprise automation [9]