NVIDIA frames Nemotron as a path to owned enterprise AI

NVIDIA is presenting Nemotron as an open model foundation for organizations that want more control over how AI systems are inspected, tuned and improved. The focus is on specialized AI, including agents and applications built for defined tasks, where models can be adapted to proprietary knowledge and evaluated against real business outcomes.

Open models are positioned as a complement to closed frontier models rather than a replacement. High-performance reasoning models can handle complex planning, while smaller customized models execute specialized tasks, helping enterprises manage inference costs, improve task-specific accuracy and adapt workflows over time without routing proprietary data through third parties.

Several companies are adapting Nemotron for industry use cases. Abridge is customizing it for clinical conversations, Glean built Waldo for enterprise search, H Company reported higher than 76% accuracy on OSWorld-Verified, and Harvey matched leading closed models on complex legal tasks at at least 10x lower cost per run. YTL AI Labs post-trained a Nemotron model for the Malaysian language.

NVIDIA also highlights supporting tools and infrastructure, including the NeMo suite for customization, evaluation, agent optimization and governance. LangChain tuned its Deep Agents harness for Nemotron 3 Ultra and reported approximately 10x lower cost per run than leading closed alternatives, while Arcee AI achieved inference costs of roughly 90 cents per million output tokens, approximately 20x cheaper than comparable closed frontier models.