Advertisement
Artificial intelligence
TechBig Tech

China’s Zhipu AI ramps up push towards human-level AI, reiterates open source pledge

The firm’s campaign to engage with the global tech community comes after it unveiled the latest iteration of its flagship GLM model series

Reading Time:2 minutes
Why you can trust SCMP
The Zhipu AI logo is displayed on a smartphone with the home page in background, June 26, 2025. Photo: CFOTO/Future Publishing via Getty Images
Ben Jiangin Beijing

Chinese AI start-up Zhipu AI said it would step up efforts next year in the pursuit of artificial general intelligence (AGI) while continuing to open source its AI models after its initial public offering.

In 2026 “we’re going to contribute more substantially to the AGI journey,” Zheng Qinkai, a researcher at the company, which markets itself internationally as Z.ai, said in an Ask Me Anything (AMA) session hosted on US social media site Reddit.

AGI, referring to technology that matches human capabilities across a wide range of domains, is a goal that AI companies around the world are racing to achieve.

Advertisement

Chinese tech giant Alibaba Group Holding, which has invested in Zhipu AI, has repeatedly stated its goal of advancing in the field of AGI. Alibaba owns the South China Morning Post.

Zhipu AI, which has just passed a regulatory hearing for a listing on the Hong Kong Stock Exchange, assured global developers that it would keep open sourcing its AI models when asked if it would continue “releasing model weights” after going public.

Zhipu AI has just passed a regulatory hearing for a listing on the Hong Kong Stock Exchange. Photo: Shutterstock Images
Zhipu AI has just passed a regulatory hearing for a listing on the Hong Kong Stock Exchange. Photo: Shutterstock Images

Model weights are numerical values learned during training that encode a model’s capabilities. It is common practice for open-source AI developers to release these, along with research papers.

Advertisement
Advertisement
Select Voice
Choose your listening speed
Get through articles 2x faster
1.25x
250 WPM
Slow
Average
Fast
1.25x