Nemotron 3 Nano 30B

Nemotron 3 Nano 30B is a production-ready, open-weight 30B-parameter Mixture-of-Experts model built on a hybrid Mamba-2 and Transformer architecture with roughly 3.5B active parameters per token. The hybrid design aims for high throughput and fast responses while keeping accuracy on par with standard transformers. The model handles mathematical reasoning, code generation, and long-context understanding, and supports tool use and reasoning modes for agentic systems. It offers 256K token context (the default Hugging Face configuration, scalable further with more memory) and works across multiple natural languages and many programming languages. Designed for developers building AI agents, retrieval-augmented generation systems, chatbots, and instruction-following applications, it suits complex problem-solving at a reduced memory footprint compared with larger dense models.