NVIDIA Nemotron Achieves Benchmark-Leading Performance With LangChain Deep Agents Harness
NVIDIA’s Nemotron 3 Ultra model, tuned with LangChain’s Deep Agents harness, delivers top-tier performance at one-tenth the inference cost of leading closed models without retraining, enabling continuous evaluation and faster agent development.
NVIDIA Nemotron 3 Ultra has achieved leading performance on LangChain’s Deep Agents benchmark while operating at a significantly lower inference cost compared to top closed models. The model maintained business task parity with the highest-scoring closed alternatives without requiring retraining, demonstrating that performance gains stemmed from system-level optimizations rather than model adjustments. LangChain’s Deep Agents platform, with over 200 million monthly downloads, was specifically tuned for Nemotron 3 Ultra to enhance task completion rates, speed, and operational efficiency.
LangChain’s engineering team focused on refining the environment around Nemotron 3 Ultra, including system prompts, tool descriptions, and middleware, to improve agent performance. The tuned harness is now available directly through LangChain, allowing developers to deploy high-performing agents immediately. This approach underscores the value of optimizing the broader system rather than altering the model itself, as noted by Harrison Chase, LangChain’s cofounder and CEO.
NVIDIA NemoClaw for LangChain Deep Agents provides an open reference blueprint that packages the tuned system for enterprises building specialized AI agents. It combines the optimized LangChain Deep Agents Code with the NVIDIA OpenShell secure runtime, enabling safe execution of agent actions. This open stack—comprising an open model, harness, and runtime—allows businesses to customize, govern, and deploy agents across their infrastructure or cloud environments without vendor lock-in.
The collaboration between NVIDIA and LangChain supports enterprises in embedding specialized agents into their platforms, with early adopters including Abridge, Amdocs, Box, and global integrator EY. NemoClaw for LangChain Deep Agents and the tuned Nemotron 3 Ultra profile are now available, offering developers direct access via LangChain or hosted platforms like Baseten, Crusoe Cloud, and Together AI. EY provides implementation support for enterprises seeking to build and govern custom agent systems.