Photo: implicator.ai

Z.ai Launches GLM-4.5: A Cost-Effective and Smarter AI Model

Beijing, July 2025 — Chinese AI startup Z.ai, formerly known as Zhipu, announced Monday the release of its latest artificial intelligence model, GLM-4.5, which promises to be more affordable and efficient than DeepSeek’s acclaimed AI system. This new model embraces an innovative “agentic” AI architecture that autonomously breaks down complex tasks into smaller, manageable components to improve accuracy and performance.

Unlike many proprietary AI models, GLM-4.5 is open source, allowing developers worldwide to download and utilize it free of charge.

Smaller, Faster, and More Affordable: Technical Highlights

GLM-4.5 is approximately half the size of DeepSeek’s model and requires only eight Nvidia H20 chips to run effectively. These GPUs were customized by Nvidia specifically for the Chinese market to comply with U.S. export regulations. Nvidia recently announced it would resume shipments to China after a three-month suspension, though the exact timeline remains uncertain.

Z.ai CEO Zhang Peng told CNBC that the company currently has sufficient computing power and declined to disclose the exact training costs of GLM-4.5, promising to reveal more details later.

For context, DeepSeek previously disclosed that its V3 model cost under $6 million to train, although some experts believe the actual hardware investment exceeds $500 million.

Competitive Pricing: A Clear Edge Over DeepSeek and Others

Z.ai’s pricing model significantly undercuts competitors:

• Input token cost: 11 cents per million tokens, compared to DeepSeek’s 14 cents
• Output token cost: 28 cents per million tokens, dramatically lower than DeepSeek’s $2.19

Tokens represent units of data processed by AI models, making these figures crucial for businesses evaluating operational expenses.

Alibaba-backed Moonshot’s Kimi K2 model, another emerging competitor, charges 15 cents per million input tokens and $2.50 per million output tokens, positioning GLM-4.5 as the most cost-efficient among recent entrants.

China’s Expanding AI Landscape and Geopolitical Tensions

Z.ai’s rise occurs amid growing scrutiny from U.S. regulators. Earlier this month, OpenAI flagged Zhipu (Z.ai’s former name) in a warning about Chinese AI advancements. The U.S. has also placed Z.ai on its entity list, restricting American companies from engaging in business with the startup.

Founded in 2019, Z.ai is preparing for an initial public offering (IPO) in Greater China. The company has raised over $1.5 billion from investors including tech giants Alibaba, Tencent, and venture firm Qiming Venture Partners, alongside Aramco-backed Prosperity7 Ventures and municipal investment funds from Hangzhou and Chengdu.

Recent Developments: China’s AI Innovation Surge

Chinese AI firms continue to push forward with open-source projects:

• At the World AI Conference in Shanghai, Tencent debuted the HunyuanWorld-1.0 model for generating 3D gaming environments.
• Last week, Alibaba introduced the Qwen3-Coder AI model, designed for programming code generation.

This momentum highlights China’s strategic drive to compete with U.S.-based AI leaders like OpenAI, especially under tightening trade and technology restrictions.

Conclusion

Z.ai’s GLM-4.5 represents a significant stride in making powerful AI models more accessible and affordable while maintaining high functionality. As global AI competition intensifies, developments like these could reshape the technology landscape, with China positioning itself as a critical player in the open-source AI ecosystem.

‍