Deepin 25 OS achieves major RISC-V breakthrough by fully adapting to Jindie Space K3 chip. This integration enables offline large model inference out-of-the-box, running the Xiao Wu AI model on small hardware. The K3 chip features a heterogeneous architecture with 8 general, 8 AI, and 2 real-time cores, delivering 60 TOPS of native AI power. Deepin optimized the toolchain and introduced the Next repository for RVA23 compliance, ensuring efficient task scheduling to AI cores. This milestone marks the first time a desktop OS on RISC-V supports offline large model inference, advancing domestic open-source software and hardware ecosystem collaboration.