NVIDIA Nemotron 3 Ultra gains tuned LangChain agent harness

NVIDIA Nemotron 3 Ultra is being positioned as a lower-cost open model option for enterprise agents after LangChain tuned its Deep Agents harness for the model. On LangChain’s Deep Agents benchmark, Nemotron 3 Ultra achieved the highest accuracy among open models, reached business task parity with the highest-scoring closed models and ran at 10x lower inference cost per run than leading closed models.

The gains came from harness engineering rather than model retraining. LangChain analyzed execution traces from Nemotron 3 Ultra and adjusted system prompts, tool descriptions and middleware around the model. The tuned profile is available directly through LangChain, whose agent engineering platform has more than 200 million monthly downloads.

NVIDIA NemoClaw for LangChain Deep Agents packages the work as an open reference blueprint for specialized AI systems, combining LangChain Deep Agents Code tuned for Nemotron 3 Ultra with NVIDIA OpenShell for secure agent action execution. Abridge, Amdocs and Box are embedding specialized agents into their platforms, while EY is expanding implementation capabilities around NVIDIA NemoClaw blueprints.

Developers can access Nemotron 3 Ultra through Baseten, Crusoe Cloud, DeepInfra, Fireworks, Nebius and Together AI. NVIDIA NemoClaw for LangChain Deep Agents and the tuned Nemotron 3 Ultra model profile are available now.