DeepSeek launches an improved version of V3 AI three months after the first release
China's DeepSeek has released an updated version of its large-scale language artificial intelligence, DeepSeek-V3-0324, just three months after the launch of the first V3 version in December 2024.
Here's What We Know
The new model has improved performance and capabilities, including the creation of aesthetically pleasing web pages and high-quality reports in Chinese.
DeepSeek-V3-0324 has attracted attention for its reduced hardware requirements for training, faster training time, and lower API costs with high performance compared to competitors such as OpenAI's GPT.
DeepSeek performance. Illustration: DeepSeek
The updated model does not involve thinking, which allows it to provide answers quickly, without lingering on complex tasks, unlike the DeepSeek R1 model. The new version has a parameter size of 685 billion, making it one of the largest publicly available language models.
DeepSeek-V3-0324 scored from 5.3% to 19.8% higher on AI tests compared to the previous version. In terms of performance, it is close to such leaders as GPT-4.5 and Claude Sonnet 3.7.
In addition, the updated version has demonstrated improvements in web page creation, as well as in searching, writing, and translating texts in Chinese.
To test the full version of the model, users will need at least 700 GB of free disk space and several Nvidia A100/H100 GPUs. However, there are also simplified versions of the model that can run on a single GPU, such as the Nvidia 3090.
Source: DeepSeek
