DeepSeek has released an update to its V3 AI model, dubbed V3-0324, which is said to offer better programming capabilities while setting new standards for accuracy and efficiency, Bloomberg reports. The update was published on the Hugging Face platform without an official announcement.
Image source: Solen Feyissa/unsplash.com
The DeepSeek V3 open AI model was released late last year. The model is built on a Mixture of Experts (MoE) architecture with a total of 671 billion parameters and 37 billion parameters activated per token.
DeepSeek V3 cost $5.5 million to train, the company said at the time, significantly less than what other tech companies like OpenAI spend on training similar models. A few weeks later, DeepSeek released an open-source reasoning model, R1, which, despite a modest development budget, outperformed OpenAI’s o1 AI model in several key benchmarks.
In January 2025, DeepSeek surpassed AI chatbot ChatGPT to become the most popular free app on the Apple App Store in the United States. The Chinese startup’s achievements raised questions among investors about the justification for American companies’ huge spending on AI technology, leading to a collapse in the AI market, with its participants seeing a significant drop in stock prices. In particular, market leader Nvidia lost $593 billion in market value in a single day, the largest one-day decline in stock market history.
Noctua plans to release its first liquid cooling system in 2026. Moreover, it will work…
CD Projekt management clarified its position on the use of generative AI in the development…
This week, Apple announced the date of its annual WWDC conference, where it will unveil…
The Chinese car market is large enough to justify deep adaptation of foreign models sold…
The decentralized platform World Liberty Financial, launched by the family of US President Donald Trump,…
Alibaba Group Chairman Joe Tsai has warned that more AI data centers may be built…