Z.ai Challenges U.S. Sanctions with New AI Models

Beijing-based Z.ai, recently added to the U.S. Entity List, has secured $1.5 billion in funding and released two open-source AI models that outperformed many Western alternatives in global benchmarks, as the company prepares for a Hong Kong listing.

Background

Previously known as Zhipu AI, the startup is now listed on Washington’s Entity List, which identifies foreign entities undermining U.S. national security. Z.ai’s recent blog post criticized OpenAI, citing its AI platform as overly complicated, while promoting its own new model, GLM-4.5.

GLM-4.5 Versus OpenAI’s Models

Z.ai boasts that the GLM-4.5 model uniquely combines reasoning and non-reasoning capabilities into a single versatile architecture. They compared OpenAI’s GPT-3, which leverages reinforcement learning for various tasks but lacks comprehensive performance across the board. Z.ai aims to unify these different capabilities with GLM-4.5.

OpenAI anticipates releasing GPT-5, aiming to be the first fully unified AI tool by late summer. In the meantime, Z.ai’s models, GLM-4.5 and GLM-4.5-Air, ranked third globally in industry tests with scores of 63.2 and 59.8 respectively, trailing OpenAI and Grok-4.

Technical Highlights

The GLM-4.5 model employs a Mixture of Experts architecture with 355 billion parameters, though only 32 billion are active at any one time for efficiency. It achieved a 90.6% tool-calling success rate, surpassing competitors in web browsing and complex problem-solving benchmarks.

Impact of U.S. Sanctions

Zhipu was placed on the Entity List in January due to alleged military collaborations. However, the company claims this designation has not significantly impacted operations; state-backed investments continued unabated. Recent funding for Z.ai includes 500 million yuan from Huafa Group and a 1 billion yuan investment from Shanghai investors.

Z.ai plans to raise $300 million through a Hong Kong IPO, shifting from its original plans for a mainland listing amid geopolitical tensions.

Cost-Effectiveness and Open Source

Z.ai’s API pricing is notably low, at $0.11 per million input tokens and $0.28 per million output tokens, substantially cheaper than Western models. They provide model weights under an MIT license for community use and modification.

Competitive Landscape in China

Chinese companies are rapidly advancing in AI development, with over 1,500 large language models released. Z.ai’s funding strategy aligns with OpenAI’s ecosystem development while maintaining open licensing. The company also launched a 1.5 billion yuan fund, Z Fund, to support startups utilizing its technology.

Conclusion

Z.ai stands at the forefront of China’s AI ambitions, striving for technological independence amidst U.S. political constraints, as demonstrated by its innovative models and robust funding strategy. Try an online version of its chatbot here.