Deepseek is the latest model compatible with the Chinese “next generation” artificial foil

Deepseek, the start of the Chinese artificial intelligence, hinted that China will soon get the domestic “next generation” chips to support artificial intelligence models, with an announcement of an update of one of its large linguistic models.

In a comment under the post on Official WeChat accountDeepseek said that the newly released “UE8M0 FP8” accuracy is designed for the upcoming chips created locally and that will be launched soon.

FP8, or 8 bit floating point, is a Data processing format The mathematical efficiency can enhance training and the conclusion of large deep learning models.

Deepseek to the next generation chips in China may indicate plans to work closely with the ecological ecosystems system in China in the face of the semiconductor export restrictions in Washington and pushing Beijing to self -sufficiency of the chip.

The comments come about two weeks after the Chinese artificial intelligence developers urged the use of local alternatives to the NVIDIA graphics processing units used to train artificial intelligence. While analysts say the Chinese artificial intelligence makers are lagging behind NVIDIA in technological progress and size, they were players like Huawei Progress.

On Thursday, Deepseek did not reveal the chips that I used to train V3.1, or what are the local chips that may be UE8M0 FP8 compatible with it.

Deepseek rocked the technology world earlier this year after she released the R1 thinking model, which showed similar capabilities for Western competitors such as Openai, despite the US export controls that restrict it from using the most advanced AI chips in NVIDIA.

Before that, in December, the company issued the V3 model, which he said was trained on about 2000 of the least advanced NVIDIA chips.

After the Deepseek model, the United States has tightened more export restrictions in April, effectively banned the NVIDIA H20 chips, which were specially designed to meet the previous export restrictions on China.

Last month, the Trump administration officials said they intend to allow NVIDIA to resume shipping chips to China. However, the H20S is now meeting with the audit in China, with the organizers Imposing companies against purchase Chips until the National Security Review is completed.

Tell CNBC chips that companies you love Huawei It may seek to build an alternative AI’s ecosystem in China that can benefit from the H20S deficiency in the market on the market.

Deepseek said on Thursday that V3.1 came with “big changes”, including the fastest response times, and the hybrid structure allows the model to support both illogical and non -seasonal conditions. Thinking models can carry out more complicated tasks through a step -by -step logical thinking.

Starting September 6, the company will also set prices to use the model programming interface, which allows other application developers and web products to integrate Deepseek on their platforms.