China’s latest AI model claims to be cheaper to use than Deepsek

Beijing-Chinese companies are creating smart artificial intelligence models that are cheaper to use, shaking in the depic market .They are echoing important things.

Startup Z.AI, who was previously known as Zipu, announced on Monday that the cost of his new GLM -4.5 AI model would be lower than using Deepsik. Unlike the logic of the existing AI models, Z.AI has said that its new GLM -4.5 “Agentic” has been created on what is known as AI, meaning the model automatically breaks the model to fulfill the model more accurately.

The new model is also open source, meaning that the developer is free to download and use.

In half of the model of Deepsek’s model, GLM -4.5 only requires eight Nvidia H20 chips, ZAI CEO Zhang Peng told CNBC on Monday to operate.

Customized for China to comply with the US export control is a chip nvidia. Chipmaker said this month that the United States will allow China to resume sales after a three -month pause, but it is unclear when the shipment will start.

Zhang said the company now does not need to buy more chips because computing power is sufficient, but the ZAI refused to share how much ZAI spent to train the AI model. The details will be announced later, he said.

In January, Deepsek had confused world investors with a clear ability to disobey American chip restrictions and create an AI model that not only compensated for ChatGPT of the US-based Open, but also reduced the training and operating costs.

Deepasik Claimed The training cost for its V3 model was less than 6 million Million, though some analysts have said that the company’s hardware costs are based on the cost of the company Over time million Million more than 500 million.

ZAI said that for your new GLM -4.5 model, it will charge 11 cents per million input tokens against 14 against R1; And 28 cents per million output tokens against $ 2.19 for Deepsec. Token is a way Counting data for AI model processing??

At the beginning of this month, Alibaba-supported Moonshot release As K2Who claimed to defeat Open ChatGPT and Anthropic Claud on some coding capabilities. Km K2 charges 15 cents for each 1 million input token and its website, charges 50 2.50 per 1 million output token.

By the end of June, An open name Zipu in the hint About Chinese AI progress. The United States has also added startup to it List of existence This prevents American companies from doing business with it.

Z.Ai is launched in 2019 and is strikingly Planning the initial public offer Greater in China.

According to the pitchbook, startup has deposited more than $ 1.5 billion from investors, including Alibaba, Tenscent and Kimming Venture Partners. Database has shown that ARAMCO-supported prosperity 7 activities as well as municipal funds in Hongzo and Chengdu cities. There is one of the backers.

Over the past few weeks, many other Chinese companies have announced a new, open-source AI model. During the World AI conference in Shanghai, Tansent released the HunuanWorld -1.0 model to create a three -dimensional look for the development of the game. Last week, Alibaba announced its Qwen3-Coder model to write a computer code.