This week, NVIDIA announced the launch of its Nemotron 3 Ultra, which has been optimized with LangChain's Deep Agents harness to achieve benchmark-leading performance on an open stack for enterprise use.
The Nemotron 3 Ultra provides high performance at a lower cost compared to leading closed models by utilizing the LangChain Deep Agents harness. This integration has resulted in the highest accuracy among open models, completing more tasks at higher throughput, and reducing inference costs by tenfold compared to top closed models. Notably, this performance gain was achieved without retraining the model itself, instead focusing on engineering the environment around the model.
LangChain's platform is widely adopted with over 200 million monthly downloads. By tuning its Deep Agents harness specifically for Nemotron 3 Ultra, it allows enterprises to deploy high-performing agents that can complete more tasks swiftly while maintaining a fully customizable and open stack that they can control and run anywhere. As Harrison Chase, cofounder and CEO of LangChain, stated, the focus is on improving system components around the model, such as memory, tool use, and evaluation.
NVIDIA's Nemotron 3 Ultra, combined with the LangChain Deep Agents harness, targets enterprises looking to build specialized AI systems. Companies like Abridge, Amdocs, and Box are already embedding these agents into their platforms, while EY is expanding its capabilities to help clients customize and govern specialized agents in high-value workflows.
Work implications: This release can enhance roles involved in AI system development and management, particularly those concerned with customizing and controlling enterprise AI solutions.
Originally reported by NVIDIA