Zhipu AI, one of China’s leading foundational model startups, has kicked off preparations for an IPO, making it the first of China’s six AI unicorns to pursue a listing, as the sector begins to diverge under competition from entrants like DeepSeek.
Spun from Tsinghua University’s Department of Computer Science in 2019, Zhipu began focusing on large language model research in 2020. It was among the first Chinese companies to explore large-scale pretraining, and is best known for co-developing GLM-130B, a bilingual Chinese-English model with over 100 billion parameters. That foundation led to the launch of open-source conversational model ChatGLM and its lightweight single-GPU variant ChatGLM-6B. Zhipu has since built a broad portfolio of AI-native products, including the productivity assistant Qingyan, the code-generation model CodeGeeX, the multimodal model CogVLM, and the text-to-image model CogView.
The company signed an IPO advisory agreement on March 31, 2025, with preparatory work beginning in April and the second phase of guidance scheduled between August and October, putting it on track to become the first foundational model startup to go public in China.
On April 15, Zhipu open-sourced its new GLM 32B/9B series under the MIT license. The release includes base, inference, and reasoning models – all available for free via its newly launched platform Z.ai and integrated into its MaaS (Model-as-a-Service) platform. According to the company, its inference model GLM-Z1-32B-0414 reaches speeds of up to 200 tokens per second, making it the fastest commercial model in China to date – and it’s priced at just one-thirtieth the cost of DeepSeek-R1. [Bloomberg, TechNode report]