China’s Z.ai Ups the AI Race With Affordable GLM-4.5 Release

China’s AI sector just notched another major milestone. Z.ai, previously known as Zhipu AI, has introduced GLM-4.5, an innovative open-source model that is garnering attention due to its exceptional performance and innovative pricing strategy.

The launch at the World AI Conference in Shanghai signals more than a new product; it marks China’s AI ambitions stepping squarely onto the global stage, with Z.ai now vying for the top spot against heavyweight rivals.

Benchmarking a New Contender

GLM-4.5’s core achievement is clear: it matches or exceeds international standards at a fraction of the cost. The model is built on a Mixture of Experts architecture, running two potent versions: the flagship with 355 billion parameters (32 billion active) and a leaner 106-billion-parameter option (12 billion active).

According to CEO Zhang Peng, GLM-4.5 achieves its output on just eight Nvidia H20 chips, slashing hardware needs compared to competitors. That efficiency translates to speedier deployments and lower total costs for enterprise clients.

Did you know?
Z.ai’s GLM-4.5 was evaluated on 12 industry benchmarks and ranked first among open-source models worldwide, despite requiring only half the hardware of key rivals.

Flexible Intelligence, Fast Responses

GLM-4.5 isn’t just about big numbers. It’s designed to switch between modes: a stepwise “thinking” process for heavy reasoning and a “non-thinking” mode for instant answers. Benchmark results back up the tech, scoring an average of 63.2 across a dozen tough industry tests and rising to the top among open-source models globally.

For developers and enterprises, its success means a toolset equipped to handle everything from complex problem-solving to high-volume, quick-turn tasks.

ALSO READ | Autonomous AI system ASI-Arch reshapes neural architecture design

Slashing Costs and Shifting the Market

What makes Z.ai stand out the most? Price. At just 11¢ per million input tokens and 28¢ per million output tokens, GLM-4.5 undercuts DeepSeek and even Moonshot’s Kimi K2 by significant margins. For context, it’s over 130 times cheaper on input and nearly 270 times on output compared to stalwarts like Claude 3 Opus.

These aggressive economics have the potential to disrupt business models and hasten global adoption. By democratizing access to sophisticated AI, Z.ai is betting that scale and openness will win out over closed, high-cost systems.

Strategic Moves and Future Ambitions

This release isn’t happening in a vacuum. As of July 2025, China leads the world with 1,509 large language models released, with Z.ai right at the front. Backed by $1.5 billion in funding from players like Alibaba, Tencent, and Shanghai’s local government, Z.ai looks ready to expand, potentially choosing Hong Kong over mainland China for a public listing.

Yet, despite momentum, the company faces geopolitical headwinds. Being on the U.S. entity list limits partnerships with American tech companies. Still, Z.ai’s leadership frames these hurdles as opportunities to double down on innovation and autonomy.

The New AI Race: Open, Fast, and Affordable

Open-source accessibility, lean hardware demands, and record-low pricing Z.ai's strategy is challenging preconceived notions about the cost and location of high-end AI development. With GLM-4.5, the company has laid down a bold challenge, not only for OpenAI and DeepSeek but for the wider AI ecosystem.

As more Chinese AI firms push to dominate this sector, global users are left with more choice, better tools, and the chance for broader collaboration. Z.ai's commitment to openness and affordability in a rapidly evolving field could pave the way for the next phase of the global AI race.