NVIDIA Nemotron 3 Ultra Leads LangChain's Deep Agents Benchmark with Tenfold Inference Cost Reduction

LangChain has optimized its Deep Agents framework for the NVIDIA Nemotron 3 Ultra, achieving the highest accuracy among open models in the LangChain Deep Agents benchmark, along with completing more tasks with higher throughput and 10x lower inference cost compared to leading closed models. The tuning did not require retraining the model: all improvements were obtained through engineering environment configuration, including system prompts, tool descriptions, and middleware. LangChain is a platform for developing agents with over 200 million downloads per month; developers can now use the tuned Nemotron 3 Ultra profile to build specialized agents. NVIDIA NemoClaw for LangChain Deep Agents is an open reference implementation that combines Deep Agents code optimized for Nemotron 3 Ultra with the NVIDIA OpenShell secure execution environment. Companies such as Abridge, Amdocs, and Box are embedding specialized agents into their platforms, and global systems integrator EY is expanding NVIDIA deployment capabilities using NemoClaw. Access to Nemotron 3 Ultra is provided through platforms such as Baseten, Crusoe Cloud, DeepInfra, Fireworks, Nebius, and Together AI.