June 22, 2026 – Zhipu AI’s latest GLM 5.2 model has claimed the third spot on the Artificial Analysis Global Leaderboard for AI intelligence, marking the first time a Chinese model has broken into the top three.
According to a new note from Jefferies, the model also ranks fourth worldwide in coding proficiency and second in agentic capabilities — results the brokerage calls a landmark moment for China’s AI trajectory.
GLM 5.2, Zhipu’s most capable open-source release to date, brings two standout improvements: it turns the previously theoretical 1-million-token context window into a production-ready feature, and it delivers stronger long-context coding performance. The model ships under the permissive MIT license, allowing free commercial use, and notably — both its training and online inference were completed without relying on overseas compute resources.
On its launch day, GLM 5.2 was already compatible with eight domestic hardware platforms, achieving full Day 0 support across Huawei Ascend, Cambricon, Moore Threads, Hygon, Biren, MetaX, Kunlun Xin, and T-Head. The “open-source Chinese model plus domestic chip” pairing has drawn significant industry attention.
Jefferies did flag valuation concerns. Assuming Zhipu hits its year-end 2026 target of $1 billion in annualized recurring revenue, the stock trades at a steep 94× P/ARR — roughly five times Anthropic’s ~18× multiple.
The brokerage also noted that U.S. access restrictions on Anthropic’s Claude 5 are unlikely to hold, leaving developers with plenty of alternatives, including open-source options.
