
NVIDIA Nemotron 3 Ultra, a 550B parameter hybrid Transformer-Mamba MoE model, is now available on Amazon SageMaker JumpStart with one-click deployment, offering 5x faster inference and up to 30% lower costs for agentic AI workloads.


<div><p>This article announces the day-zero availability of NVIDIA Nemotron 3 Ultra on Amazon SageMaker JumpStart, a new open language model optimized for agentic AI workloads.</p><ul><li>550B total parameters with 55B active parameters using hybrid Transformer-Mamba MoE architecture</li><li>5x faster inference and up to 30% lower cost for agentic workloads</li><li>Supports up to 1M token context length for long-running autonomous agents</li><li>One-click deployment through SageMaker JumpStart with no infrastructure management required</li><li>Ideal for agent orchestrators, coding agents, deep research, and complex enterprise workflows</li><li>Requires GPU instances like ml.p5en.48xlarge with associated hourly costs</li></ul><p>Nemotron 3 Ultra enables efficient multi-step reasoning for production agents by activating only necessary parameters, maintaining coherence across hundreds of turns while managing costs effectively.</p></div>


Related articles