China could have its own 'Mythos-class' AI sooner than expected, says Zai chief after GLM 5.2 launch

The global artificial intelligence race may be entering a new phase. Just weeks after the US tightened restrictions on access to some of the world's most advanced AI models, a leading Chinese AI executive has suggested that China could soon develop comparable systems of its own.

Tang Jie, founder of Chinese AI lab Zai, made the comments following the launch of GLM 5.2, the company's latest large language model. His remarks have attracted attention because they come amid growing competition between the United States and China over advanced AI capabilities.

The comments also arrived after recent restrictions on access to Anthropic's most powerful AI models for foreign users, reportedly due to concerns that hostile actors could misuse them for cyber operations.

A disagreement with Elon Musk

The discussion began on X when a user suggested that China could have a "Mythos-class" AI model by the end of 2026. Elon Musk responded by predicting that such a development was more likely around the first quarter of 2027.

Tang disagreed.

"Won't take that long," he replied, signalling confidence that Chinese AI development may be progressing faster than many observers expect. The exchange quickly drew attention from AI researchers and technology analysts because it touched on one of the industry's most important questions: how quickly China can close the gap with leading American AI labs.

Why GLM 5.2 is generating attention

Much of the discussion centres on GLM 5.2 itself. According to benchmark data cited by supporters of the model, GLM 5.2 has become one of the highest-performing open-weight AI models currently available. Open-weight models make their learned parameters publicly available, allowing developers and researchers to run, study and modify them locally. GLM 5.2 reportedly achieved a score of 51 on the Artificial Analysis Intelligence Index. That places it ahead of several well-known models, including Google's Gemini 3.5 Flash and Anthropic's Claude Sonnet 4.6 on that benchmark. However, it still trails newer frontier systems such as Anthropic's Fable 5 and OpenAI's GPT 5.5.

Trained without Nvidia chips

One detail attracting particular interest is how the model was built. According to Zai, GLM 5.2 was trained entirely using Huawei Ascend chips rather than Nvidia hardware. That matters because US export restrictions have significantly limited China's access to some of Nvidia's most advanced AI processors. The success of a high-performing model trained on domestic hardware could be viewed as evidence that Chinese companies are finding alternative paths forward despite restrictions. Reports have also suggested the training costs were substantially lower than those associated with many leading Western AI models.

The debate extends beyond benchmark scores. Advanced AI systems are increasingly viewed as strategic technologies with implications for cybersecurity, economic competitiveness and national security. American policymakers have argued that limiting access to cutting-edge AI systems can help reduce certain risks, including the misuse of powerful models for cyberattacks. However, if Chinese firms develop comparable capabilities independently, the long-term impact of those restrictions could become less clear. At the same time, leading US companies are unlikely to stand still.

Anthropic, OpenAI, Google and other major labs continue to release more capable systems at a rapid pace, meaning the benchmark for frontier AI is constantly moving higher.

The next chapter in the US-China AI competition

The exchange between Tang Jie and Elon Musk highlights how quickly the AI landscape is changing. Only a few years ago, the world's most advanced AI systems were concentrated among a handful of American companies. Today, Chinese firms are producing increasingly competitive models and attracting growing international attention. Whether China can develop a true frontier model within the timeline suggested by Tang remains uncertain. What is clear, however, is that the competition between the world's two largest AI ecosystems is accelerating, and the next major breakthrough may arrive sooner than many expected.