R2 from DeepSick: The Chinese model that may outperform GPT-4o

Amid growing global anticipation, speculations are accelerating about the next model of open-source artificial intelligence developed by the Chinese company "DeepSea," known as "R2."

This highly anticipated model follows the great success of the company's "R1" model, which caused a stir in Silicon Valley earlier this year.

Leaks and discussions on social media platforms suggest that "R2" will outperform in terms of efficiency and cost, thanks to its use of hybrid "expert mix" engineering and containing 1.2 trillion parameters, making it around 97% cheaper than the American model "GPT_4o."

It has been trained on servers using local "Huawei" chips, marking a significant shift towards reducing reliance on American supply chains.

Although the company has not confirmed the launch date, observers expect "R2" to be unveiled this month. So far, "DeepSea" remains silent, content with releasing improvements to its previous model "V3," adding more mystery to the project.

In light of this data, "R2" is seen as a strategic step that enhances China's position in the global artificial intelligence race and imposes a new reality in competition with Western technology companies.