China Will Have a Mythos-Class AI Model by Year-End, Says Z.ai Founder

China may be much closer to achieving frontier AI capabilities than many industry observers expect, according to Jie Tang, Founder of Chinese AI startup Z.ai.

SpaceX CEO Elon Musk has forecasted that top Chinese large language models could reach the performance level of Anthropic’s newly released Claude Fable 5 by the first quarter of 2027.  To which, Zhipu AI Co-Founder Tang Jie immediately pushed back, stating that Chinese models will get there sooner.

The exchange took place on X after a user asked about the timeline for Chinese AI models to achieve ‘Fable-class’ capabilities, citing the recent release of Z.ai's GLM-5.2.

The user argued that GLM-5.2 had already narrowed the gap with leading Western models, placing it roughly at the level of Claude Opus 4.7–4.8 with an estimated seven-month lag. Based on that assessment, the user suggested that China could reach ‘Fable’ status by late 2026.

“Possibly in the first quarter of 2027,” Musk responded. 

Tang pushed back on that timeline, replying, “It won’t take that long.”

Anthropic recently unveiled Claude Fable 5 and the more powerful Claude Mythos 5, setting new standards in coding, reasoning, science, and agentic tasks. Days later, US export restrictions forced the company to withdraw the models from international markets.

In response to this, Z.ai launched GLM-5.2, its latest flagship model built for long-horizon coding, software engineering, and autonomous agent workflows.

The company claims the model delivers major improvements in agentic reasoning and complex engineering tasks, while supporting a context window of up to one million tokens.

According to Z.ai, GLM-5.2 is designed specifically for long-horizon tasks where AI systems must plan, reason, and execute over extended periods rather than generate isolated responses. The model introduces multiple reasoning modes and builds on the GLM-5 family launched earlier this year.

The release has intensified discussion around China’s progress in frontier AI. 

According to benchmark results shared by Z.ai, GLM-5.2 outperforms OpenAI's GPT-5.5 on several coding and engineering evaluations, suggesting that Chinese AI labs are narrowing the gap with their US rivals.

Founded by computer scientist Jie Tang and originally known as Zhipu AI, Z.ai is widely regarded as one of China’s leading AI Tigers—a group of startups competing with firms such as DeepSeek, MiniMax, Moonshot AI, and Western leaders including OpenAI, Anthropic, and Google. 

The company rebranded globally as Z.ai in 2025 and has increasingly focused on open-source releases and AI agents.

The company’s recent model releases have arrived at a rapid pace. GLM-5 debuted in February with a focus on agentic engineering and coding tasks, followed by GLM-5.1 in April, which expanded autonomous software engineering capabilities.