NVIDIA's Nemotron 3 Ultra has achieved the highest accuracy among open models on LangChain's Deep Agents benchmark while running at 10x lower inference cost than leading closed models.

LangChain tuned its Deep Agents harness specifically for Nemotron 3 Ultra, delivering business task parity with top closed models without any model retraining. The performance gains came entirely from engineering the environment around the model — adjusting system prompts, tool descriptions and middleware.

"The way to build better agents is to keep improving the system around the model," said Harrison Chase, cofounder and CEO of LangChain. "Memory, tool use, evaluation and model behaviour compound when teams can tune them together."

The collaboration produces higher throughput and faster task completion on an open stack that enterprises can run, customise and control. LangChain's agent platform sees more than 200 million monthly downloads.

Open Stack for Enterprise Control

NVIDIA NemoClaw for LangChain Deep Agents packages the work as an open reference blueprint for enterprises building specialised AI systems. It combines the tuned LangChain Deep Agents code with NVIDIA OpenShell secure runtime for safe agent execution.

The open model, harness and runtime give enterprises full stack ownership. They can customise around their business expertise, iterate continuously and deploy on their own infrastructure with their own governance.

Abridge, Amdocs and Box are embedding specialised agents into their platforms. Global systems integrator EY is expanding NVIDIA implementation capabilities around the NemoClaw blueprints, helping clients customise and govern agents across high-value workflows.

Developers can access Nemotron 3 Ultra through Nebius, Baseten, Crusoe Cloud, DeepInfra, Fireworks and Together AI platforms. The tuned Deep Agents harness is available directly through LangChain today.

NVIDIA founder Jensen Huang recently discussed with Chase why the past six months have delivered a leap in useful enterprise AI. The shift from AI assistants answering questions to agents taking action inside core systems changes what businesses extract from their AI investments.