The China Depseek startup, which caused shock waves in all markets this year, silently launched an improved version of its artificial intelligence model.
The company did not make an official announcement, but Deepseek R1 update was launched in the AI models repository that embraces the face.
Depseek jumped to fame this year after its free R1 Reasoning model and open source surpassed the offers higher than the rivals, including Goal and Openai. The low cost and the short time of development shocked global markets, generating Conns that US technological giants were spending too much on infrastructure and cleaning billions of dollars of value of technological value of the United States as an unconditional nvidia of AI. These companies have recovered in general terms.
As was the case of Deepseek R1 debut, the updated model was also launched with little fanfare. It is a refueling model, which means that AI can execute more complicated tasks through a step -by -step logical thinking process.
The Deepseek R1 updated model is only O4-MINI and O3 OPENI Reasoning Models in LiveCodeBench, a site that shares models with different metrics.
Deepseek has become the son of the poster of how Chinese artificial intelligence is still developing despite the United States attempts to restrict access to the country to fried potatoes and other technologies. This month, Chinese technology giants Baidu And Tencent revealed how their AI models are more efficient to deal with the export curbs of US semiconductors.
Jensen Huang, CEO or NvidiaHe designs the graphics processing units necessary to train huge AI models, criticized US export controls on Wednesday.
“The United States has based its policy that China cannot do AI chips,” said Huang. “That assumption was always questionable, and now it is clearly bad.”
“The question is not if China will have AI,” Huang added. “He already does.”