DeepSeek V4 Pro and Harness Launch on National Supercomputing Internet with 100k GPUs

Official release of DeepSeek V4 Pro and open-source Harness framework on national supercomputing network, leveraging 100,000 GPUs for full lifecycle model support.

Woofun AI reports that the National Supercomputing Internet has deployed the official version of DeepSeek V4 Pro alongside the DeepSeek Harness Agent framework. The platform utilizes China's first superintelligence integration computing resource pool, comprising 100,000 GPUs, to deliver full lifecycle hash rate support for large models to research institutions, tech startups, and developers.

DeepSeek-V4-Pro-0813 prioritizes enhanced Agent capabilities. The concurrently released DeepSeek Harness is open-sourced under the MIT license, employing a 'plugins for everything' architecture that enables flexible replacement and recombination of Agent components including models, tools, skills, and conversations. The framework supports four operating modes: Standard, PTC, Minimalist, and Creative.