Hundreds of millions of dollars! The director of Ant AGI Research Center started a large model business and received new financing

Zhidongxi Author | Li Shuiqing Editor | Yunpeng Zhidongzhi reported on August 3 that just now, Hangzhou Large Model Company Xihu Xinchen announced that the company has recently completed several hundred million yuan in Series B+ financing. This round of financing was led by GF Xinde, with participation from West Lake Innovation Investment, Wuhan Finless Porpoise Fund, Rongyi Investment, Nantong Investment Management and other institutions. Industrial capital such as 58 Industry Fund participated. Old shareholders Ant Group and Tom Cat provided full support. Financing funds will be mainly invested in three directions: continuous research and development of a new generation of diffusion language models, in-depth iteration of its native multi-modal real-time interactive products, and further expansion of overseas markets. Xihu Xinchen was established in July 2021, headquartered in Hangzhou, Zhejiang, and established based on Westlake University. The company has long been focusing on the research and development of native multi-modal large models under the diffusion paradigm, and takes real-time interactive content as the core implementation scenario. The founder of the company, Lan Zhenzhong, is a PhD in AI from Carnegie Mellon University and the first author of the natural language processing pre-trained language model "ALBERT". He was a scientist at the Google Artificial Intelligence Research Institute. His research results have been applied to hundreds of millions of user products such as Google News and Google Assistant. He was employed by Westlake University in June 2020, founded the deep learning laboratory and served as a doctoral supervisor. He also serves as the director of the General Artificial Intelligence Center of the Ant Group Technology Research Institute. ▲ Lan Zhenzhong, founder of Xihu Xinchen. According to public information, Xihu Xinchen has completed angel round, Pre-A round, two rounds of strategic financing, round B, and round B+, and a total of 6 rounds of public financing. The total disclosed financing scale has reached hundreds of millions of yuan. The current mainstream paradigm of large models is the autoregressive architecture, which is characterized by word-by-word serial generation, but there is a ceiling in efficiency and controllability. Xihu Xinchen believes that the diffusion language model adopts an iterative generation path of "from noise to order" and has structural advantages such as parallel generation, native editability, cross-modal unification and stronger controllability. In 2026, the technological progress of diffusion language models will accelerate significantly. Google releases Gemma Di