A massive mixture-of-experts model that activates only 17 billion of its 397 billion parameters per forward pass, keeping inference costs manageable while drawing on a vast pool of learned representations. It handles text-in, text-out tasks with the efficiency characteristic of sparse architectures. The NVFP4 V2 quantization from NVIDIA trades a small amount of numerical precision for significantly reduced memory footprint, making deployment on high-end hardware more practical.