NVIDIA Nemotron 3 Ultra, tuned for LangChain's Deep Agents platform, matches the performance of premium closed-source mo

NVIDIA Nemotron 3 Ultra is delivering leading performance at significantly lower cost compared to top closed models, particularly when integrated with LangChain's Deep Agents orchestration platform, which has over 200 million monthly downloads.

LangChain optimized its Deep Agents harness specifically for Nemotron 3 Ultra and achieved the highest accuracy among open models. The tuned system completes more tasks at higher throughput while running at ten times lower inference cost per run than leading closed models. Crucially, no model retraining was required to reach this performance level. Every improvement came from engineering the environment around the model through adjustments to system prompts, tool descriptions and middleware.

The tuned Nemotron 3 Ultra model profile demonstrates business task parity with the highest-scoring closed models on LangChain's Deep Agents benchmark. At one-tenth of the cost, teams can run evaluations continuously, experiment more rapidly and build specialized agents across broader business operations.

Harrison Chase, LangChain's cofounder and CEO, emphasized that enterprise success depends on continuous improvement of the system surrounding the model. He noted that memory, tool use, evaluation and model behavior compound when teams can tune them together, and demonstrated that enterprises can achieve strong performance from an open stack while maintaining full control over their agent systems.

Several organizations are already deploying this technology. Abridge, Amdocs and Box are embedding specialized agents into their platforms. Global systems integrator EY is expanding its NVIDIA implementation capabilities around NVIDIA NemoClaw blueprints for LangChain Deep Agents, helping clients customize, evaluate and govern specialized agents across high-value workflows.

NVIDIA NemoClaw for LangChain Deep Agents serves as an open reference blueprint combining LangChain Deep Agents code, tuned for Nemotron 3 Ultra, with the NVIDIA OpenShell secure runtime for safely executing agent actions. This fully open stack approach means enterprises own the entire end-to-end system, can customize it around their unique business expertise, continue improving it and deploy it anywhere on their own infrastructure and governance.

The tuned Nemotron 3 Ultra model profile and NemoClaw blueprint are available immediately. Developers can access the tuned Deep Agents harness directly from LangChain or use the NemoClaw blueprint as a starting point for building specialized agents. LangChain developers can also access Nemotron 3 Ultra on Baseten, Crusoe Cloud, DeepInfra, Fireworks, Nebius and Together AI platforms for hosted production deployment. EY is positioned to help enterprises begin building specialized agents today using this open software stack.