Core Insights - The core task of AI infrastructure is shifting from training and inference of large models to a new phase focused on "intelligent circulation," emphasizing the efficient and stable use of model capabilities in real business scenarios [1] - The market demand for model API services is emerging as the industry transitions to a stage where the focus is on how to operate models long-term, stably, and at scale [1] Company Overview - Qingcheng Jizhi was founded in December 2023 in Beijing, with a core founding team from Tsinghua University's Computer Science Department [1] - The company has previously launched the "Bagua Furnace" training system and the "Chitu" inference engine to support efficient training and deployment of models across various computing powers [1] Product Development - The AI Ping platform, launched by Qingcheng Jizhi, offers a one-stop AI evaluation and API service routing platform, aimed at enhancing the infrastructure capabilities during the application phase of large models [1] - AI Ping covers a complete chain from "evaluation—access—routing—optimization," focusing on real business scenarios and providing long-term observation of key metrics such as latency, stability, throughput, and cost-effectiveness for different model APIs [2] Market Analysis - A joint report with Huqing Puzhi titled "Large Model API Service Industry Analysis Report (2025)" indicates that DeepSeek and Qwen series models dominate the open-source model API calls, with significant performance differences among service providers [3] - The report highlights that the average daily consumption of large models in China's enterprise market reached 10.2 trillion tokens in the first half of last year, indicating a shift from seeking a single strongest model to finding optimal solutions for specific business scenarios [3]

Group 1 - The core viewpoint of the articles is the introduction of the AI Ping platform by Qingcheng Jizhi, which focuses on the evaluation, unified access, and intelligent routing of large model services, creating a complete link covering "evaluation - access - routing - optimization" [1][3] - AI Ping currently covers over 30 domestic large model API service providers, providing comparative analysis of model service capabilities under unified standards and methodologies, aiding enterprises in making rational decisions amidst complex model and service choices [1] - The shift in the core tasks of AI infrastructure is highlighted, moving from training and inference of large models to a new stage focused on "intelligent circulation," emphasizing the efficient and stable use of model capabilities in real business scenarios [2] Group 2 - The key to achieving intelligent circulation lies in the construction of intelligent routing capabilities, which includes model routing for selecting the most suitable model for different tasks and service routing for optimizing performance and cost among various API service providers [2] - Qingcheng Jizhi's CEO, Tang Xiongchao, emphasizes the evolving focus of AI infrastructure from training and inference to the higher demands for service stability and efficiency during the application phase, leading to the development of AI Ping as a one-stop AI evaluation and API service intelligent routing platform [3] - The development of a complete AI task distribution network through the collaborative development of model and service routing capabilities is crucial for determining the final efficiency and cost of artificial intelligence systems [2]