China’s Z.ai Debuts GLM-4.5, Challenges OpenAI’s Strategy

Backed by $1.5 billion in new funding, the company appears unaffected by the restriction and is also preparing to list its shares on the Hong Kong stock exchange.

The company, previously called Zhipu AI, was placed on the US Entity List in January. This limits its access to American-made technology unless special permission is granted.

Did you know?

Want to get smarter & wealthier with crypto?

Subscribe - We publish new crypto explainer videos every week!

What is a Cryptocurrency: For Beginners (Animated Explainer)

SUBSCRIBE

ON YOUTUBE

Z.ai has released GLM-4.5 and a lighter version called GLM-4.5-Air. These models are open-source and designed to handle multiple tasks simultaneously, which utilizes only a portion of their full capacity during each process.

In tests commonly used to measure AI performance, GLM-4.5 earned a score of 63.2, while the smaller version scored 59.8. These results placed them third overall, behind only OpenAI’s o3 and Grok-4.

One test involved calling external tools to complete tasks. Z.ai’s main model had a 90.6% success rate, higher than Claude 3.5 Sonnet (89.5%) and Kimi K2 (86.2%). In answering hard questions using web search, it got 26.4% right, beating Claude 4 Opus, which managed 18.8%.

Z.ai also used this launch to question the approach of OpenAI. The company said OpenAI’s tools are becoming too specific and scattered. Z.ai noted that its GLM-4.5 system can handle a range of tasks, from reasoning to basic knowledge, all in one model.

Recently, OpenAI launched a new ChatGPT feature called "agent mode" for its Plus, Pro, and Team subscribers. What does it do? Read the full story.