Sibling line: July 2026 nemotron-3-embed open embedding family tops rteb (nvidia-nemotron-3-embed-rteb).
NVIDIA’s 550B-parameter (55B active) open-weight MoE model optimized for long-running agentic AI workloads, announced at GTC Taipei / Computex 2026.
Recent Developments
- 2026-06-01: Unveiled at Jensen Huang keynote; ships June 4 on Hugging Face, OpenRouter, build.nvidia.com
- Claims 5x faster inference, 30% lower cost, 300+ tokens/sec throughput
- Scores 48 on Artificial Analysis Intelligence Index — leading US open models, trailing China’s Kimi K2.6 (54)
- Bundled with free NVIDIA Agent Toolkit (NemoClaw, OpenShell, CUDA-X skills)