Z.ai cuts costs with new open-source AI model, underpricing DeepSeek
Chinese startup Z.ai launched its new AI model GLM-4.5, promising cheaper AI use than DeepSeek. The model runs on “agentic” AI, breaking tasks into sub-tasks automatically for better accuracy.
GLM-4.5 is about half the size of DeepSeek’s model and only needs eight Nvidia H20 chips — the Nvidia chip customized for China amid US export controls. Z.ai CEO Zhang Peng told CNBC on Monday the company has enough chips for now and will reveal training costs later. 
“The new model is also open sourced, meaning it is free for developers to download and use.”
Z.ai charges 11 cents per million input tokens, versus DeepSeek’s 14 cents. Output tokens cost 28 cents per million, compared to DeepSeek’s $2.19.
The launch follows DeepSeek’s January bombshell, which rattled global investors by rivaling OpenAI’s ChatGPT while undercutting training and operating costs. DeepSeek claimed training its V3 model cost under $6 million, though analysts questioned that figure given its $500 million-plus hardware spend.
Alibaba-backed Moonshot also jumped in earlier this month with Kimi K2, claiming better coding skills than OpenAI and Anthropic, at a higher token price.
Z.ai, formerly Zhipu, was flagged by OpenAI in June as a key Chinese AI threat and added to the US entity list restricting American business ties. The startup has raised over $1.5 billion, including from Alibaba, Tencent, and Aramco-backed Prosperity7.
Recent weeks have seen other Chinese players announce new open-source models: Tencent released HunyuanWorld-1.0 for 3D game scenes, and Alibaba launched Qwen3-Coder for code generation, both at the World AI Conference in Shanghai.
“Zhang said the company doesn’t need to buy more of the chips as it has enough computing power for now, but declined to share how much Z.ai spent on training the AI model. Details will be released later,” CNBC reported.
Z.ai is also reportedly planning an IPO in Greater China, continuing its rapid rise despite US restrictions.
