The national team end! DeepSeek all models landing on the national supercomputing platform

Recently, DeepSeek-R1, V3, Coder and other complete series of large models under DeepSeek were officially connected to the national supercomputing Internet platform (www.scnet.cn), and integrated computing power in domestic superintelligence and national integrated computing power services. With the support of the system, China's AI model industry has taken a key step.

The National Supercomputing Internet Platform is currently launched by 1.5B, 7B, 8B, and 14B of the DeepSeek-R1 model. Users can directly experience online reasoning services by scanning code or web. No local deployment is required. In response to enterprise -level needs, the platform supports the introduction of proprietary data for privatization model custom development. Large -parameter versions such as 32B and 70B will also be opened in the near future. Simultaneous launch also includes vertical field models such as DeepSeek-V3, Coder series, Math series (7B), covering diverse scenarios such as code generation and mathematical reasoning.

Compared with large overseas models, the DeepSeek series has significantly reduced computing resources through algorithm optimization, combined with the national supercomputing platform "inclusive and easy -to -use, rich and easy to use" computing power scheduling capabilities to achieve great compression of calling costs. Take DeepSeek-R1-Distill-Qwen-7B as an example. Its WebUI dialogue interface supports the zero threshold experience, and enterprise users can quickly access the production environment through the API interface to form a complete link from model calls to privatization deployment. Essence

Analysts pointed out that the scale of the DeepSeek model on the supercomputing platform may change the current status of the domestic AI industry to rely on overseas closed sources. With more developers based on domestic computing power base training and optimizing localized models, China is expected to achieve batch "curve overtaking" on the AI ​​technology application layer. AI competition injected new variables.