“Won’t take that long.” That was Tang Jie’s response after businessman Elon Musk posted that Chinese artificial intelligence companies were still years away from matching the world’s leading AI models. Less than two weeks later, Mr. Tang, the 49-year-old co-founder of Zhipu AI, set out to back up his claim
On June 16, Zhipu AI (Z.ai), listed in Hong Kong as Knowledge Atlas Technology, unveiled GLM-5.2, its most advanced large language model yet, alongside ZCode, an AI coding platform designed to write, edit and manage software. Early results suggest the model is now competing closely with Anthropic’s latest systems, including Mythos, adding momentum to China’s AI race with the U.S. The timing could hardly have been more significant. While American companies continue to dominate the frontier of AI, the U.S. has tightened export controls on advanced chips and AI technologies in an effort to slow Beijing’s progress. China, meanwhile, has doubled down on home-grown innovation. Over the past year, companies such as DeepSeek, Alibaba, Baidu and Moonshot AI have unveiled increasingly capable models. Z.ai is now seeking to join the list.
Born in China, Tang Jie, also a professor of computer science at Tsinghua University, co-founded Zhipu AI in 2019 as a university spin-off and today serves as its chief scientist. Before launching the company, Mr. Tang was best known for creating AMiner, an academic search engine and knowledge graph platform used by researchers around the world
Since its founding, Z.ai has steadily released successive generations of its GLM (Generative Pre-trained Language Models) family of language models, with each version edging closer to the capabilities of leading American systems. GLM-5.2 is the company’s biggest leap so far. Researchers say GLM-5.2 is among the first to seriously challenge leading U.S. systems in specialised coding and cybersecurity tasks
In tests conducted by cybersecurity firm Semgrep, GLM-5.2 outperformed Anthropic’s Claude Opus 4.8 on certain software engineering tasks. In other evaluations, researchers found it could match the bug-finding capabilities of Anthropic’s frontier models. The model has also climbed into the top 10 most-used AI systems on OpenRouter, a platform that provides developers access to hundreds of AI models
‘DeepSeek moment’
The launch has drawn comparisons with DeepSeek, the Chinese startup whose breakthrough model shook the global AI industry last year. According to South China Morning Post, some developers in Silicon Valley have described GLM-5.2 as another ‘DeepSeek moment’ for China’s open-affordability
According to The Wall Street Journal, unlike OpenAI’s ChatGPT or Anthropic’s Claude, GLM-5.2 has been released as an open-weight model. In simple terms, developers can download, modify and run it on their own hardware rather than relying on Z.ai’s servers. Supporters say that makes cutting-edge AI significantly cheaper and more accessible. Critics, on the other hand, argue the same openness could allow hackers and malicious actors to deploy AI systems without the many safeguards imposed by commercial providers.
Alongside the model, Z.ai has also introduced ZCode, a coding assistant that can write, edit and manage software projects with minimal human input. Together, the company says, the two products are designed to make AI a more capable partner for software developers
Yet, for all the excitement, GLM-5.2 is not without its critics. Researchers note that it still trails OpenAI and Anthropic on several complex reasoning and general-purpose tasks. Others point to the constraints Chinese AI companies continue to face because of U.S. restrictions on access to advanced semiconductor hardware
According to Z.ai, however, the model delivers near-frontier performance at a fraction of the price charged by comparable proprietary systems
While the competition between the U.S. and China intensifies, the company has sought to strike a collaborative tone. “Competition and collaboration are what push all of us forward,” Zixuan Li, Z.ai’s head of global operations, wrote in a post on X following the launch
