Nemotron

Nemotron is a family of highly efficient, multimodal, open AI models from NVIDIA, built for long-running, self-evolving agents. The models are designed for fast task completion with high reasoning throughput and accuracy across complex agent workflows.

The reasoning models come in three tiers: Nano (superior accuracy and efficiency for specialized sub-agents), Super (highest-accuracy, high-throughput reasoning with tool calling for multi-agent systems) and Ultra (best reasoning for mission-critical applications). Nemotron handles diverse agentic AI workloads: reasoning, multimodal vision, retrieval-augmented generation (RAG), speech (ASR, TTS, neural machine translation) and safety filtering; it excels in computer-use agents, document intelligence, and video and audio understanding.

The models are truly open — NVIDIA publishes the training datasets, techniques and model weights under the permissive NVIDIA Open Model License (use, modify, distribute and commercially deploy without crediting NVIDIA). They are available for free on Hugging Face, via NVIDIA NIM microservices (with NVIDIA AI Enterprise licensing) and through build.nvidia.com and inference service providers.