Zhipu AI Founder: China Will Match Anthropic's Top Models by 2026
Taylor Wilson
Zhipu AI founder Tang Jie publicly countered Elon Musk, saying China can match Anthropic's flagship Claude Fable 5 without waiting until Q1 next year — a rare open exchange that drags the China-U.S. AI gap debate from vague narrative onto a concrete timeline.
What exactly was said in this exchange?
It started with an X post asking when China could field a Fable 5 competitor — Anthropic's flagship, widely regarded as the world's most advanced AI model.
Musk commented "probably Q1 next year." Tang Jie, Zhipu AI's founder, replied it would not take that long — pulling the timeline into 2026.
This means → the head of a leading Chinese AI lab has, for the first time, made a public, time-bound commitment to matching the world's strongest model.
Where does Tang's confidence come from?
Zhipu AI just released GLM-5.2, a 74.4-billion-parameter model that ranks second globally in front-end coding on Code Arena's benchmark — behind only Fable 5.
That result pushed Zhipu AI to third place on the Artificial Analysis Intelligence Index, after Anthropic and OpenAI.
In plain terms = Zhipu is not making empty claims — its latest model already trails Fable 5 by a narrow margin on one hard benchmark, and that is the data behind "it won't take that long."
Does ranking second in coding equal "matching" Fable 5?
It does not. Code Arena's front-end coding ranking captures one slice of AI capability; Fable 5 earns its top status from reasoning, multimodal, and long-context performance combined.
This means → GLM-5.2's coding result proves the gap is narrowing, but validating Tang's claim requires sustained gains across a broader set of benchmarks.
This reflects a shift in the China-U.S. AI race — from "who has more parameters" to "who scores higher on key benchmarks." Concrete scores now matter more than broad narratives.
What does this mean for the market?
If Tang's call holds, China will have a Fable 5-class model within 2026 — a development that would reshape the competitive landscape of the global AI supply chain.
The verification window is short: GLM-5.2's performance on benchmarks beyond coding, and Zhipu's iteration speed on follow-up models, will deliver answers within the next few quarters.
In plain terms = this is the "talk" phase; the "proof" phase comes next — benchmark data will be far more telling than any exchange on social media.
Content is for reference only, not financial advice.