China’s National Supercomputing Internet has officially launched DeepSeek V4 Pro alongside the DeepSeek Harness agent framework. This move integrates advanced AI capabilities with a large-scale computing infrastructure to support users throughout the full life cycle of model development. The platform now hosts a 100,000-accelerator resource pool designed for training and deploying large-scale AI models.
The DeepSeek Harness framework uses a modular plugin architecture. This allows developers to swap components like models and tools to suit different tasks. It operates in four distinct modes: Standard, PTC, Minimalist, and Creative. By releasing the source code under an MIT license, the platform enables smaller technology firms and individual developers to build and customize their own AI agents with lower technical barriers.
Experts note that this development signals a shift in AI application. Instead of models that merely provide answers, these new agents act as workers that plan workflows and execute specific tasks. This expansion covers areas from software engineering to industrial simulation and scientific research. The platform currently supports over 1,700 open-source models, reinforcing the local ecosystem for domestic AI development.