National Supercomputing Network Launches DeepSeek V4 Pro and Harness

China's IT Home reports that the National Supercomputing Network has officially launched the DeepSeek V4 Pro artificial intelligence model and the DeepSeek Harness development framework. The announcement was made on August 14.

The National Supercomputing Network is described as a large-scale computing resource pool designed to support AI model development. The platform aims to provide research institutions, technology companies, and developers with capabilities for training, fine-tuning, evaluating, and deploying large language models. It reportedly enables users to download DeepSeek models, perform private deployments, and develop intelligent agents.

The DeepSeek V4 Pro model is noted for its enhanced agent capabilities, with performance in production environments claimed to be comparable to closed-source foreign models such as Claude Fable 5 and Opus-4.8.

A developer preview version of the DeepSeek Harness framework was also released for testing on August 13, with its source code made available under an MIT license. Users are expected to be able to download project files directly from the platform's code repository for further development and deployment.