In the world of modern information technologies, the struggle to create the smartest AI model has long gone beyond Silicon Valley. Chinese developers are actively catching up with Western giants, and in this race, DeepSeek has announced a new version of its powerful R1 model posted on the Hugging Face platformwhere developers from all over the world can freely test it.
What’s new in the DeepSeek R1 model
DeepSeek announced the update in an official post on WeChat. According to the company, this is a minor update, but still important — the model has become more accurate in complex logic tasks that require deep understanding and analysis. The model has 685 billion parameters, which is a truly gigantic amount of data that puts it on par with the most powerful AI systems in the world. For comparison, the model OpenAI GPT-3 has 175 billion parameters. Without modifications, the updated model will probably not be able to run on consumer-level computer hardware.
The new version of DeepSeek R1 was released under the MIT open source license, so it can be used even in commercial projects without legal restrictions. Interestingly, the Hugging Face repository contains only configuration files and weights — these are the internal numerical parameters that actually form the «brain» model. In other words, the documentation is still minimal, but enthusiasts can already start experimenting.
The Chinese startup became known at the beginning of the year when presented the R1 model. It showed impressive results and made even experts who had previously focused exclusively on OpenAI products talk about themselves. This rapid growth has caused concern in some government agencies in the United States. They believe that such developments can pose a risk to national security, as they can be used not only for peaceful purposes.
Source: techcrunch
Spelling error report
The following text will be sent to our editors: