A massive mixture-of-experts model from NVIDIA with 550 billion total parameters but only 55 billion active at a time, giving it a wide knowledge base without the full computational cost of dense models. It handles an exceptionally long context window of 262,144 tokens, making it capable of reasoning over very large documents or codebases in a single pass. As an open-weight release, it can be run locally or fine-tuned, though its sheer scale means serious hardware is required.