nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16 · Hugging Face
NVIDIA has released Nemotron-3-Ultra-550B-A55B-BF16, a large language model designed for advanced agentic capabilities and long-context analysis. The model features a hybrid Latent Mixture-of-Experts architecture with Mamba-2 and Attention layers, supporting up to 1 million tokens. It is optimized for complex reasoning, tool use, and multilingual tasks, with a total of 550 billion parameters and 55 billion active parameters. AI
IMPACT Sets new SOTA for agentic reasoning and long-context analysis, potentially influencing future specialized AI development.