An artificial intelligence assistant from the Chinese startup DeepSeek has for the first time surpassed ChatGPT and became the most popular free program in the App Store in the USA, according to Sensor Tower (via Reuters).
DeepSeek V3 is an open-source model that has performed quite well in key tests and has shown that it can compete with leading industry models, while also showing “improvements” in logical reasoning tasks. V3 uses a Mixture-of-Experts (MoE) architecture with a total number of parameters at 671 billion, where 37 billion are activated for each token (in the predecessor V2, for comparison, there were 236 billion parameters with 21 billion parameters activated for each token).
Interestingly, DeepSeek V3 used “a cluster of over 2000 Nvidia chips” — compared to the tens of thousands of chips for training similarly sized models, which American companies spend millions of dollars on.
This is also the same model that considered itself ChatGPT — here’s why this is happening.
In recent years, Chinese tech giants and startups have released a whole range of AI models, but DeepSeek has managed to capture attention worldwide and cause panic in the offices of Silicon Valley. Recently, the company released the reasoning model R1 in open access, prompting OpenAI to move its o3-mini model into the free version of ChatGPT.
R1 also has its own records — it became the most popular model downloaded on HuggingFace (over 100,000 downloads as of January 25). Also, a few days ago, independent analysts from Chatbot Arena (a platform organized by researchers at the University of California, Berkeley) added the V3 and R1 models to the top ten most productive chatbots.
Spelling error report
The following text will be sent to our editors: