DeepSeek-V4-Flash official API now on national supercomputing platform

China's national supercomputing internet platform has launched the official version of DeepSeek-V4-Flash (DeepSeek-V4-Flash-0731), offering API access and model files. After extensive post-training, the model shows big improvements in agent capabilities and instruction following, performing on par with top closed-source models in benchmarks.

Enterprises and developers can now visit the platform's website, go to the model services section, and get API access with one click—no complex setup needed. The platform's AI community has gathered over 1,700 popular open-source models, including DeepSeek, GLM, Qwen, and Kimi, with APIs for DeepSeek-V4, MiniMax-M3, and Kimi-K3 already live.

With its core nodes now operational, the platform has become China's first supercomputing resource pool with 100,000 GPUs, supporting large-scale AI applications.