DeepSeek launches V4 Pro flagship model alongside local AI cluster innovations

DeepSeek has officially launched its flagship 1.6-trillion-parameter model, DeepSeek V4 Pro 0813, moving it from preview to general availability. The release includes a new open-source agent framework and model weights released under the MIT License. The update focuses on significant improvements in agentic performance, with benchmark scores showing substantial gains in areas such as software engineering (DeepSWE) and cybersecurity (CyberGym) compared to previous preview builds.

In a separate development regarding local AI deployment, GEEKOM has demonstrated a four-node AI cluster using its A9 Mega Mini PCs. The cluster utilizes AMD Ryzen AI Max+ 395 processors to run DeepSeek V4 Flash locally. By connecting the units via USB4, the system provides a distributed platform for businesses to run advanced AI models on private infrastructure, allowing for secure processing of sensitive documents and code without relying on public cloud services.