NVIDIA Nemotron 3 Ultra, optimized for LangChain's Deep Agents platform, matches the performance of expensive closed AI

NVIDIA and LangChain have partnered to deliver open-source AI agents that compete with the highest-performing closed models on cost and capability. LangChain, whose agent platform has over 200 million monthly downloads, tuned its Deep Agents harness specifically for NVIDIA Nemotron 3 Ultra and achieved the highest accuracy among open models while completing more tasks at higher throughput and running at a tenth of the inference cost per run compared to leading closed alternatives.

The key innovation was engineering rather than retraining. LangChain's team analyzed where their agents lost performance points and adjusted system prompts, tool descriptions and middleware around the base Nemotron 3 Ultra model. This approach achieved business task parity with the highest-scoring closed models without any model retraining, proving that significant gains come from optimizing the environment around the model.

At this lower cost, enterprises can run continuous evaluations, experiment faster and deploy specialized agents across more workflows. Harrison Chase, cofounder and CEO of LangChain, emphasized that improvement comes from tuning memory, tool use, evaluation and model behavior together, allowing enterprises to maintain full control over their open-stack agent systems.

NVIDIA introduced NemoClaw for LangChain Deep Agents, an open reference blueprint that combines LangChain's tuned Deep Agents code with NVIDIA's OpenShell secure runtime for safe agent execution. This fully open stack lets enterprises customize agents to their specific workflows, improve them continuously and run them anywhere on their own infrastructure or cloud.

Early adopters including Abridge, Amdocs and Box are embedding specialized agents into their platforms, while global systems integrator EY is expanding its implementation capabilities to help clients build, evaluate and govern specialized agents. Developers can access the tuned Nemotron 3 Ultra profile directly through LangChain or use the NemoClaw blueprint as a starting point, with hosted access available through Baseten, Crusoe Cloud, DeepInfra, Fireworks, Nebius and Together AI.