The NVIDIA Nemotron 3 Ultra stands as the largest and most widely adopted AI agent orchestration platform, offering superior performance at a significantly reduced cost compared to leading proprietary models.
LangChain has optimized its Deep Agents harness specifically for the NVIDIA Nemotron 3 Ultra, ensuring unmatched accuracy in open models while efficiently completing more tasks with increased throughput—executing at a 10x lower inference cost per run than top closed models.
In tests applying LangChain’s Deep Agents benchmark, the Nemotron 3 Ultra performed business tasks on par with the leading closed model without the need for model retraining; all advancements derived from engineering the model environment rather than the model itself.
Utilizing the NVIDIA Nemotron 3 Ultra, teams can conduct continuous assessments, accelerate experimentation, and develop expert agents across diverse industries at just one-tenth the cost.
LangChain’s agent engineering platform sees over 200 million downloads monthly. Through the targeted adjustments of the Deep Agents harness, NVIDIA Nemotron 3 Ultra empowers high-performance agents capable of accomplishing more tasks swiftly, providing businesses with a completely open stack that they can personalize, control, and operate anywhere.
“To enhance agent performance, we continually refine our systems based on our models,” stated Harrison Chase, co-founder and CEO of LangChain. “Memory, tool usage, evaluation, and model behavior become intricate when teams collaborate on tuning. Our partnership with NVIDIA demonstrates that organizations can achieve exceptional performance from open stacks while maintaining governance over their agent systems.”
Companies like Abridge, Amdocs, and Box integrate expert agents directly into their platforms, while global system integrators like EY enhance NVIDIA’s implementation capabilities around the NVIDIA NemoClaw Blueprint for LangChain Deep Agents, aiding clients in customizing, evaluating, and managing specialized agents across high-value workflows.
Recently, NVIDIA’s Founder and CEO Jensen Huang engaged with Chase to explore the remarkable advancements in AI for enterprises over the past six months.
Engineering Over Fine-Tuning
The LangChain team assessed the Nemotron 3 Ultra using the available Deep Agents benchmark suite. Deep Agent execution traces reveal areas for improvement. Instead of retraining models, the team optimized the harness, allowing for tailored system prompts, tool descriptions, and middleware adjustments.
All developers utilizing LangChain Deep Agents on Nemotron 3 Ultra can immediately benefit from these enhancements, with targeted profiles readily available through LangChain.
An Open Stack for Ownership
The NVIDIA NemoClaw for LangChain Deep Agents serves as an open reference blueprint, facilitating companies in their own development efforts. Specialized AI, tools, and runtimes are tailored to fit unique workflows. LangChain Deep Agents code optimized for Nemotron 3 Ultra, combined with NVIDIA Open Shell ensures a secure environment for executing agent actions.
With open models, harnesses, and secure runtimes, enterprises retain full ownership of their end-to-end stack. Continually refine processes based on unique business expertise and deploy across any infrastructure—your own cloud, and governance.
This distinction becomes increasingly significant as agents undertake riskier tasks. The transition from AI assistants responding to inquiries to agents executing actions in core systems will transform corporate outcomes from AI.
The NemoClaw and optimized Nemotron 3 Ultra model profiles for LangChain Deep Agents are accessible now. Developers can directly obtain tailored Deep Agent harnesses from LangChain or NemoClaw for LangChain. Deep Agent Blueprints serve as starting points for creating specialized agents from the ground up.
Getting Started
LangChain developers have access to the Nemotron 3 Ultra. Platforms like Base Ten, Crusoe Cloud, Deep Infrastructure, Firework, Nevius, and AI Together offer direct, hosted pathways to production-tuned harnesses.
EY supports businesses in leveraging this open software stack to develop proprietary agents.
Learn more about NVIDIA NemoClaw and NVIDIA Nemotron for LangChain Deep Agent.
Stay updated on Agent AI, NVIDIA Nemotron. Subscribe for more NVIDIA News, join the community, and follow NVIDIA AI on LinkedIn, Instagram, Twitter, and Facebook.
Source: blogs.nvidia.com