- Hangzhou firm pushes dual-track strategy targeting enterprises and small businesses
- Approval adds to growing list of registered generative AI services in China
OpenPie (拓数派), a Hangzhou-based startup specializing in trusted data space and agentic AI, said its self-developed Hansi large language model has received regulatory approval from China’s cyberspace authorities, allowing the firm to formally roll out generative AI services to corporate and individual users.
According to a filing notice released May 19 by Zhejiang’s cyberspace regulator, the Hansi model completed China’s generative AI service registration process on May 15 under filing number “ZheJiang-TuoShuPaiHanSiDaMoXing-202604130084.”
The approval marks the latest addition to China’s growing roster of officially registered AI models. The Cyberspace Administration of China said 868 generative AI services had completed filing procedures nationwide as of April 30.
Also on May 19, Manycore Tech, another Hangzhou-based AI company, said its spatial intelligence model — focused on AI’s physical-world parsing and recognition as well as related applications — had also received official approval from China’s cyberspace authorities.
OpenPie focuses on cloud-native data computing systems and develops enterprise infrastructure designed to consolidate fragmented data warehouses into unified cloud platforms through its PieCloudDB virtual data warehouse product.
The company positions itself among a group of firms including Databricks and Palantir Technologies that are exploring the convergence of data-space architecture and agentic AI technologies.
The company said the Hansi model, debuted at the 2025 World Internet Conference Wuzhen Summit, has been optimized for multi-turn dialogue, structured information extraction, summarization, rewriting, content expansion, classification and scoring tasks.
Applications include AI-assisted knowledge search, industry information retrieval, travel planning and content generation.
Ray Von, founder, CEO and chairman of Tencent-backed OpenPie, previously described the Hansi model as “not just a language model, but an intelligent security steward for enterprise private data,” adding that it is designed to unlock business value while ensuring data remains “usable but invisible.”
Before securing approval for the Hansi model, OpenPie had already obtained regulatory clearance in January 2026 for its “technology text generation algorithm” under China’s deep synthesis algorithm filing system.
The company said the dual approvals support its strategy of targeting both large institutional clients and smaller businesses.
Under that structure, an enterprise-focused version is designed for government and corporate customers requiring high-concurrency AI computing support, while a professional version targets small businesses and so-called “one-person companies” through lighter-weight deployment systems and shared public token access.
Going forward, OpenPie said it plans to continue integrating large language models and trusted data infrastructure, while accelerating commercialization of low-code and no-code AI agents across industries.
