NVIDIA Nemotron 3 Ultra, tuned for LangChain's Deep Agents platform, now achieves performance matching leading closed models while running at 10x lower inference cost per run. The breakthrough came from engineering the harness around the model rather than retraining it, allowing enterprises to deploy high-performing agents they fully own and control across their own infrastructure.
Summaries like this, in your inbox every morning.
Sign up free →What happened
LangChain tuned its Deep Agents orchestration platform for NVIDIA Nemotron 3 Ultra and achieved the highest accuracy among open models on LangChain's Deep Agents benchmark, with business task parity to leading closed models—without retraining the model itself. The gains came from engineering the environment (system prompts, tool descriptions, middleware) around the model.
Why it matters
Enterprises can now run AI agents at a tenth of the cost of leading closed models, enabling teams to run evaluations continuously and experiment faster while maintaining full control over an open stack they can customize, own and run anywhere. Companies like Abridge, Amdocs, Box and EY are already embedding specialized agents into their platforms.
What to watch
The tuned Nemotron 3 Ultra profile is available now directly through LangChain, and developers can access the model on Baseten, Crusoe Cloud, DeepInfra, Fireworks, Nebius and Together AI. The NVIDIA NemoClaw for LangChain Deep Agents blueprint packages this work as an open reference for enterprises building their own specialized AI systems.
The news centers on a shift in how enterprises can deploy AI agents cost-effectively. LangChain's Deep Agents platform, which has more than 200 million monthly downloads, analyzed execution traces from Nemotron 3 Ultra to identify performance gaps, then tuned the harness rather than the model weights themselves. This engineering-first approach has resonated with early adopters: system integrator EY is expanding its NVIDIA implementation capabilities around NemoClaw blueprints to help clients customize and govern specialized agents, while platform companies Abridge, Amdocs and Box are embedding these agents directly into their systems.
The competitive dynamic here is meaningful for enterprises weighing closed versus open AI models. By achieving business task parity with leading closed models at a tenth of the inference cost, Nemotron 3 Ultra reduces a key financial barrier to open-model adoption. The emphasis on owning the full stack—open model, open harness, open secure runtime—addresses a real governance concern as agents move from answering questions to taking actions inside core business systems. Enterprises can now customize, improve and audit their AI systems without vendor lock-in, which may shape purchasing decisions for high-stakes workflows.
No comments yet. Be the first to share your thoughts!
Log in to join the discussion
Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.
Get Started FreeFree · takes 30 seconds · unsubscribe anytime
1 minute a day. The AI essentials.
200+ sources · Email / LINE / Slack
