All terms
Models & Products
Nemotron
NVIDIA's family of open-weight language models tuned for efficient, long-running agents.
Definition
Nemotron is NVIDIA's family of open-weight language models, released in a range of sizes and built to run agentic workloads efficiently rather than chase raw benchmark size alone. Newer Nemotron models use a hybrid design that mixes Mamba-style state-space layers with regular Transformer attention plus a mixture-of-experts feedforward network, so only a fraction of the model activates per request. This combination is aimed at cutting the cost of very long contexts and multi-step agent reasoning while keeping the weights open for anyone to download and run.