POPA ReconNVIDIA Unveils Nemotron 3 Ultra, a 550B Open-Weight Model for Agentic AI
NVIDIA Unveils Nemotron 3 Ultra, a 550B Open-Weight Model for Agentic AI NVIDIA has released Nemotron 3 Ultra, a 550B-parameter open-weight model designed for long-context, agentic workflows such as coding, research, and multi-step enterprise automation. The model uses a hybrid Mamba2 + Transformer + Mixture-of-Experts architecture with only 55B active parameters, enabling up to 5x faster inference and ~30% lower cost for complex agent tasks. NVIDIA also released a full open stack, including base models, post-trained versions, reward checkpoints, quantized NVFP4 weights, training data, and…
The excerpt is quoted from a public Telegram channel, which owns its text.