The latest AI Model in China claims they are even cheaper in use than Deepseek

The World Artificial Intelligence Conference began in Shanghai on Saturday, July 26, 2025.

Nurphoto | Nurphoto | Gets the image

Beijing-Chinese companies make smarter artificial intelligence models that are increasingly cheaper in use, repeating the key aspects of the Deepseek market breakthrough.

Startup Z.AI, which was previously known as Zipu, announced on Monday that its new AI GLM-4.5 model would cost less than in the depths of the series. Unlike the logic underlying the existing AI models, Z.AI said its new GLM-4.5 was built on what is known as “Agentic” AI, that is, the model automatically destroys the task at the podzoda to complete it more accurately.

The new model is also open, which means that developers are free to download and use.

Approximately half-size Deepseek, GLM-4.5 requires only eight Nvidia H20 chips for work, Z.AI CEO Zhang Pen said on Monday.

This is a Nvidia chip set up for China to match US export control. This month, the chip manufacturer said the US would allow it to resume sales in China after a three -month pause, but it is unclear when these deliveries begin.

Zhang said the companies did not need to buy more chips because it has enough computing, but refused to share how much Z.aii spent on training AI. The details will be published later, he said.

AI Trump's strategy realizes that

Back in January, Deepseek clattering global investors, with the obvious ability to challenge US chips and create a AI model, which not only sacked on Chatgpt Openai, which is based in the US, but also undermining it in training and exploitation.

Depth state Training costs for its model V3 amounted to less than $ 6 million, although some analysts said the indicator is based on the company’s equipment More than $ 500 million over time.

Z.AI said that for her new GLM-4.5 model, she charges 11 cents per million entities rather than 14 per Deepseek R1; and 28 cents per million output tokens against $ 2.19 per Deepseek. Tokens – this is a way Measurement data to process AI model.

Earlier this month released Alibaba Moonshot As k2who claimed to have won the Chatgpt Openai and the anthropic Claude of certain coding opportunities. Kimi K2 charges 15 cents for every 1 million entry characters and $ 2.50 per 1 million production, according to its site.

Weekly analysis and understanding of Asia’s largest economy in your mailbox
Sign up now

In late June, Openai called Zipu in the warning About Chinese progress II. US also added a startup to its List of essence This limits US companies to run a business with it.

Z.AI launched in 2019 and reportedly reported Planning the initial public offer in great China.

The startup has raised more than $ 1.5 billion from investors, including Alibaba, Tencent and Qiming Venture Partners, Pitchbook reports. Aramko supported by Aramco, as well as municipal funds from the cities of Hangzhou and Chengdu, are also among the fans of Z.AI, showed in the database.

In the last few weeks, several other Chinese companies have announced new open source AI models. During the AI World Conference in Shanghai, Tencent released the HunyuanWorld-1.0 model to create three-dimensional game development scenes. Last week Alibaba announced its QWEN3-code model to write computer code.

Source link