American AI coding agents are impressive. But so are China’s

There’s a new Chinese model gaining traction among developers: Zhipu AI’s GLM 4.7. This isn’t new per se, but what caught our attention was this: Where is becoming popular: USA is our territory.

I first saw Zhipu’s post on WeChat stating that the new coding tool was in so much demand that he would start limiting access. A year after DeepSeek’s R1 model shook up the AI ​​landscape in the United States, AI experts say they are seeing Chinese AI models spreading around the world. I wanted to see if this was also true for Zhipu.

Not only did DeepSeek never respond to our requests, I reached out to the company with no expectation of hearing back. But Zhipu recently went public in Hong Kong and its investor relations team is on the job. They responded almost immediately and told me that “Zhipu GLM Coding Plan’s user base is mainly concentrated in the United States and China.”

American developers have always told us about the bias against using Chinese models; so Zhipu gaining traction here suggested a true DeepSeek-like breakout moment.

Just last week we were amazed by the applications Replit and Claude Code could produce in minutes. They felt like the frontier of artificial intelligence, an example of how American innovation is leading the way. It also creates tangible change and 60% increase in new application versions. But if a model from China is this powerful and easy to use, is it really six months behind? Google What does DeepMind’s Demis Hassabis say? And what are the real moats of American AI coding agents if they are cheaper and open source?

CNBC’s Deirdre Bosa and I took the opportunity to experiment with Zhipu to see if the excitement could match the successes we’ve had with Replit and Claude. I’ve always found it difficult to navigate and get data on Chinese stock markets, so we asked him to create a tracker for the largest publicly traded companies in China. Zhipu launched an app faster than its American counterparts, but its results were not as stellar.

Naturally, we’re skeptical of claims from companies touting usage without publishing metrics, so we checked some sources — people developing AI would be up against Zhipu if they’re actually making headway here. Although reviews were mixed, it quickly became clear that Zhipu’s GLM 4.7 was indeed a model that was becoming well-known in the United States.

One developer testing this is Tuhin Srivastava from Baseten, a platform that underpins AI applications and sees real-world model use across organizations. This is even more important as Baseten embarks on a new tour. Nvidia Its inclusion is a signal that it is committed to real, large-scale AI workloads.

Watch the live broadcast with Srivastava and Bosa Here It starts at 2:30pm ET on Monday.