Nemotron 3.5 and 3 Ultra NVIDIA releases

NVIDIA released Nemotron 3 Ultra, a 550B total parameter (55B active) open-weight Mixture-of-Experts hybrid Mamba-Transformer model with 1M-token context window and ~6x higher inference throughput than comparable open LLMs. The release includes Nemotron 3.5 variants for content safety and automatic speech recognition, with Nemotron 3 Ultra available on Amazon SageMaker JumpStart offering 5x faster inference and 30% lower cost for agentic AI workloads.