A mid-sized reasoning model running in a quantized NVFP4 format optimized for NVIDIA DGX Spark hardware. It handles text-in, text-out tasks with the efficiency trade-offs typical of aggressive quantization — reduced memory footprint at the potential cost of some precision compared to full-precision variants. Part of Alibaba's Qwen 3 family, it brings multilingual capability and instruction-following to constrained deployment environments.