Nvidia Debuts Nemotron 3 Family of Open Models
21 hours ago
- #AI
- #NVIDIA
- #Machine Learning
- NVIDIA announced the Nemotron 3 family of open models, data, and libraries for agentic AI development.
- Nemotron 3 includes Nano, Super, and Ultra models with hybrid latent mixture-of-experts (MoE) architecture.
- Nemotron 3 Nano offers 4x higher throughput than Nemotron 2 Nano and is optimized for efficiency.
- Nemotron 3 Super and Ultra provide high-accuracy reasoning for complex AI applications.
- NVIDIA released training datasets and reinforcement learning libraries for AI agent customization.
- Nemotron 3 Nano is available now, while Super and Ultra models are expected in early 2026.
- Early adopters include Accenture, ServiceNow, Perplexity, and others across various industries.
- Nemotron supports sovereign AI efforts with open, transparent models aligned to regional regulations.
- The models are available on Hugging Face, AWS, Google Cloud, and other platforms.