On July 31, DeepSeek announced that the official version of the DeepSeek-V4-Flash API had entered public beta. The National Supercomputing Internet officially launched the model API calling service and model files for the official version of DeepSeek-V4-Flash (DeepSeek-V4-Flash-0731).
According to the introduction, DeepSeek-V4-Flash-0731 has undergone extensive post-training, significantly enhancing its agent capabilities and instruction-following capabilities. In multiple benchmark tests, the official version of DeepSeek-V4-Flash performed comparably to the strongest closed-source models currently available.
The Supercomputing Internet now provides enterprises and developers with API calling services for the DeepSeek-V4-Flash-0731 model. Users do not need to go through complicated environment configuration; they can quickly make calls with one-click integration. Users can log in to the PC version of the National Supercomputing Internet's official website and enter the “Model Services” section through the “Services” drop-down menu on the homepage to go directly to the model API calling page.
At present, the Supercomputing Internet AI community has brought together more than 1,700 popular open-source large models from China and abroad, covering mainstream series such as DeepSeek, GLM, Qwen, and Kimi. It has fully launched API interfaces for multiple domestic open-source large models, including DeepSeek-V4, MiniMax-M3, and Kimi-K3.
With its core nodes now officially operational, the National Supercomputing Internet has become the country's first supercomputing-intelligent computing integrated resource pool at the 100,000-card scale.