In January, Chinese AI startup DeepSeek sent global stock markets into a tailspin when it released a reasoning model called R1 that outperformed many of its competitors despite being trained at a fraction of the cost. Now the company is rushing to release its successor.
Image source: Solen Feyissa / unsplash.com
DeepSeek initially planned to release R2 in early May, but is now aiming to do so sooner rather than later, Reuters reported, citing two sources who declined to elaborate. The company hopes the new model will be able to write better code and reason in languages other than English.
DeepSeek’s competitors are still trying to recover from the fallout from the release of R1, an AI model that was trained on underpowered Nvidia accelerators but was able to compete with its counterparts trained by American tech giants for colossal sums.
The release of R2 will also be a cause for concern for the US government, which has identified the country’s leadership in AI as a national priority. On the other hand, the Chinese government and companies that have begun integrating DeepSeek models into their products will become even more active, including Lenovo, Baidu, and Tencent.