NVIDIA debuts Nemotron 3 open models for agentic AIs
The Nemotron 3 family includes Nano, Super and Ultra models plus datasets and libraries for agentic AI.
NVIDIA has unveiled the Nemotron 3 family of open models, datasets and libraries for building agentic AI systems.
The Nemotron 3 lineup — Nano, Super and Ultra — uses a hybrid latent mixture-of-experts architecture to support multi-agent workflows, the company said in an announcement.
Nemotron 3 Nano is a 30-billion-parameter model that can activate up to 3 billion parameters per token and offers a 1-million-token context window, it explained.
The chip-maker highlighted that Nemotron 3 Nano delivers 4x higher throughput than Nemotron 2 Nano and can reduce reasoning-token generation by up to 60%, lowering inference costs.
“Open innovation is the foundation of AI progress,” said Jensen Huang, founder and CEO of NVIDIA.
Nemotron 3 Super is approximately a 100-billion-parameter model with up to 10 billion active parameters per token, and Nemotron 3 Ultra is about a 500-billion-parameter engine with up to 50 billion active parameters per token, the company said.
Nvidia added that Super and Ultra use the ultra-efficient 4-bit NVFP4 training format on the NVIDIA Blackwell architecture to cut memory needs and speed training.
Artificial Analysis ranked the model as the most open and efficient among models of the same size, with leading accuracy.
NVIDIA said it is releasing three trillion tokens of pretraining, post-training and reinforcement learning datasets, plus the Nemotron Agentic Safety Dataset, and open-source libraries NeMo Gym, NeMo RL and NeMo Evaluator on GitHub and Hugging Face.
Nemotron 3 Nano is available on Hugging Face and through inference providers including Baseten, DeepInfra, Fireworks, FriendliAI, OpenRouter and Together AI, and is offered as an NVIDIA NIM microservice for deployment on NVIDIA-accelerated infrastructure.
Nemotron is supported on multiple enterprise platforms and cloud providers, and Nemotron 3 Super and Ultra are expected to be available in the first half of 2026.
The Recap
- NVIDIA launched Nemotron 3 open models, data and libraries.
- Nemotron 3 Nano delivers up to four times higher throughput.
- Nemotron 3 Super and Ultra expected in first half of 2026.
