There’s a new model from China that’s been a hot topic among developers. Zhipu AI’s GLM 4.7. It’s not new per se, but what caught our attention is where it’s gaining popularity: our turf, the United States.
The first thing I came across was Zhipu’s post on WeChat announcing that the new coding tools were in such high demand that they would start restricting access. A year after DeepSeek’s R1 model shocked the U.S. artificial intelligence industry, AI experts say Chinese AI models are spreading around the world. I wanted to know if that applies to Zhipu as well.
DeepSeek never responded to our requests, so I contacted them without expecting a response. However, Zhipu recently went public in Hong Kong and is led by an investor relations team. They replied almost immediately and said that “Zhipu GLMcoding Plan’s user base is mainly concentrated in the United States and China.”
American developers have always told us about their prejudice against the use of Chinese models. So Zhipu’s rise to prominence here suggests a true DeepSeek-like breakout moment.
Just last week, we were amazed by the apps that Replit and Claude Code were able to create in minutes. We felt they were at the frontier of AI and an example of how American innovation is leading the way. We’ve also seen a visible shift, with new app releases surging by 60%. But if the Chinese-made model is just as powerful and easy to use, are we really six months behind? google Demis Hassabis from DeepMind said: And if they are cheaper and open source, what kind of moat do American AI coding agencies really have?
CNBC’s Deirdre Bosa and I jumped at the chance to experiment with Zhipu to see if the hype could match the success with Replit and Claude. I’ve always found China’s stock exchanges difficult to navigate and retrieve data from, so I asked the company to build a tracker for China’s largest listed companies. Although Zhipu developed its app more quickly than its American competitors, the results were less sophisticated.
We are understandably skeptical of claims by companies that advertise usage without publicly disclosing metrics. So we checked with some sources, people who are building the AI that you would encounter if Zhipu were to actually make inroads. It quickly became clear that Zhipu’s GLM 4.7 is certainly a model that is gaining recognition in the United States, albeit with mixed reviews.
One builder testing this is Tuhin Srivastava of Baseten. Baseten is a platform that sits under AI applications and looks at real-world model usage across the enterprise. That’s even more important because Baseten just raised a new round. Nvidia Participation means being part of a real-world, large-scale AI workload.
Mr. Srivastava’s conversation with Mr. Botha was livestreamed here on Monday at 2:30 pm ET.
