China’s Z.ai Unveils Reasoning AI Model That's Cheaper Than DeepSeek
GLM‑4.5 is optimised to parse complex, multi-step requests with minimal human supervision.
Chinese AI startup Z.ai, formerly Zhipu, has launched its new open-source language model GLM‑4.5, claiming it is even cheaper to use than DeepSeek’s AI, intensifying the AI price war in China while grabbing global attention.
The startup claims that GLM‑4.5 is both more cost-effective and more efficient than DeepSeek’s model, with pricing set at $0.11 per million input tokens and $0.28 for output—significantly undercutting DeepSeek’s respective rates of $0.14 and $2.19.
Built to support agentic workflows—autonomous task execution—GLM‑4.5 is optimised to parse complex, multi-step requests with minimal human supervision. It is released under a permissive open-source license, making it freely available on platforms like Hugging Face and via API.
The debut follows DeepSeek’s disruptive entry earlier this year, where its DeepSeek‑V3 model demonstrated high-performance capabilities at ultra-low cost—reportedly training for just $6 million in under two months. Now, Z.ai is challenging that benchmark, offering similar or better functionality at a fraction of the operating cost.
This development highlights the escalating competition among Chinese tech players such as Alibaba, Baidu, and Tencent, each racing to build high-performance AI at low cost.
