On June 22, Business Insider reported that a new AI model from China has once again sparked heated discussions in Silicon Valley. More than a year ago, the R1 large model launched by DeepSeek shocked Silicon Valley and was regarded as a serious threat to the hegemony of American chatbots.


GLM 5.2 large model of Zhipu

Recently, online discussion circles in Silicon Valley have been abuzz with Zhipu’s new open source model. The model, called GLM 5.2, is a large language model designed for long code tasks and agent workflows.

Zhipu said the model supports context windows of up to 1 million tokens, which would put it in the same echelon as Anthropic’s Claude Opus 4.8 and OpenAI’s GPT 5.5.

"I am sincerely impressed and almost shocked by the excellent performance of GLM-5.2 in programming. This will change the landscape." Guillermo Rauch, CEO of Vercel, a cloud platform for developers, wrote on X.


Rauch praises the programming capabilities of GLM-5.2

On social media, investors, founders and tech influencers have praised the new model’s speed and performance. The model went live last week.

Mat Velloso, a former vice president at Meta, Google DeepMind and Microsoft, said on X that he spent the entire day using GLM-5.2.

“This is the first open source model that can be used as a daily workhorse,” he wrote. “The industry landscape is about to change.”


Veloso said GLM-5.2 can be used as a main model

Like DeepSeek, GLM-5.2 is open source, which means anyone can download the model, run it on their own system, and modify it as needed.

In contrast, most U.S. cutting-edge models, such as those from OpenAI and Anthropic, are typically closed source. In closed-source model mode, users must rely on the model provider. This model is better for providers because it captures more value. This is critical for companies that are investing billions of dollars in AI infrastructure and where investors have high expectations for revenue growth.

However, if an open source model performs as well or better than a closed source model, it could easily capture a larger market share.

In recent years, China and the United States have been competing fiercely for AI dominance. The United States attempts to maintain its lead through chip export restrictions and controls on computing power and technology access, while Chinese technology companies continue to advance development, launch open source models with lower costs and increasingly powerful capabilities, and continue to catch up.

Tesla CEO Elon Musk also recently talked about the narrowing of the model gap between China and the United States. Last week, some netizens asked Musk on X when China's large models are expected to reach the level of Anthropic Fable, and mentioned that Zhipu GLM-5.2 has narrowed the gap.


Musk's estimate

In response, Musk replied: "Maybe in the first quarter of 2027." However, Tang Jie, the founder of Zhipu, quickly responded that "it doesn't take that long."

In January last year, when DeepSeek released R1, a low-cost inference model comparable to OpenAI’s o1, China first sounded the alarm to Silicon Valley. At the time, investors were questioning whether Silicon Valley’s leadership in AI was as solid as it seemed.

Now, as GLM-5.2 sparks heated discussions on the Internet, the same questions have surfaced again.