China’s Z.ai Launches GLM‑4.5 at Lower Cost Than DeepSeek

China’s Z.ai Launches GLM‑4.5 at Lower Cost Than DeepSeek

Chinese model GLM-4.5 challenges DeepSeek: cheaper, more compact, open

Chinese startup Z.ai (formerly known as Zhipu AI) has unveiled a new AI model, GLM-4.5 , claiming that it is not only more productive but also significantly cheaper to operate than DeepSeek, another high-profile project in the Chinese LLM market.

GLM-4.5 is an open-source model that can be freely downloaded, deployed, and used in your own solutions. According to information posted on the company's official resources, it runs on just eight Nvidia H20 accelerators , making it one of the most affordable advanced Chinese models in 2025.

What makes the GLM-4.5 special

The developers claim that the model is built on the architecture of agent AI : unlike classic transformers, it can break down complex tasks into subtasks and perform them sequentially - which brings it closer to the logic of "smart agents". This direction is actively developing in China and is supported not only by private investors, but also at the level of industry strategies.

According to the official Z.ai blog (July 2025), the cost of processing 1 million input tokens in GLM-4.5 is $0.11 , and the output is $0.28 . This is significantly lower than the estimated cost of similar operations in DeepSeek R1, where the price for output tokens exceeds $2.

By comparison, according to Hugging Face and GitHub (where both models are hosted), DeepSeek requires roughly twice as much hardware and has a more complex architecture. This means that GLM-4.5 is potentially cheaper to run not only in the context of the API, but also for companies running the model on their own infrastructure.

China’s Z.ai Launches GLM‑4.5 at Lower Cost Than DeepSeek

China's AI Strategy and the US Price Challenge

GLM-4.5 continues a general trend in China: lowering the barriers to entry into advanced AI technologies. Recall that in January 2025, DeepSeek R1 became a sensation thanks to its ability to do without the American A100 and H100 - bypassing sanctions. Now GLM-4.5 demonstrates that it is possible to create a commercially viable LLM without super-expensive chips.

Founded in 2019, Z.ai has raised over $1.5 billion in funding, including from Alibaba , Tencent , and Qiming Venture Partners (PitchBook data, July 2025). However, due to restrictions imposed by the United States (see the US Department of Commerce Entity List), Z.ai has limited access to cutting-edge chips from Nvidia and AMD, making its success especially significant.

What's next?

Z.ai plans to promote GLM-4.5 as an industry standard for companies that require on-premises, secure AI infrastructure. Integration of the model into business products (including fintech, education, analytics) promises to be rapid — especially given that many Chinese enterprises are already testing the model.
In the long term, Z.ai could become the flagship of the so-called low-cost LLM economy, which focuses not on scaling parameters but on optimizing performance per unit of resource. And if such models can compete with GPT-4 and Gemini not in quality, then at least in economics, this will change the global AI services market.

 

By Claire Whitmore 
July 29, 2025

Join us. Our Telegram: @forexturnkey
All to the point, no ads. A channel that doesn't tire you out, but pumps you up.
FX24

Author’s Posts