NVIDIA has opened a new chapter in agentic artificial intelligence with the launch of Nemotron 3, a family of open models, datasets, and tools designed to enable transparent, efficient, and scalable multi-agent systems. The release reflects NVIDIA’s push to support enterprises moving beyond standalone chatbots toward coordinated AI agents that collaborate across complex workflows.
Built on a hybrid latent mixture-of-experts architecture, Nemotron 3 is engineered to lower inference costs, reduce context drift, and improve coordination among AI agents. These capabilities are increasingly critical as organisations deploy multiple agents to handle planning, reasoning, execution, and monitoring tasks within the same system. By optimising how models share and route information, NVIDIA aims to make large-scale agentic systems more predictable and cost-effective for enterprise use.
At the launch, NVIDIA founder and CEO Jensen Huang highlighted the company’s emphasis on openness and developer empowerment. “Open innovation is the foundation of AI progress,” he said. “With Nemotron, we’re transforming advanced AI into an open platform that gives developers the transparency and efficiency they need to build agentic systems at scale.”
The Nemotron 3 lineup includes Nano, Super, and Ultra variants, offering flexibility across performance and cost requirements. The portfolio ranges from a 30-billion-parameter model optimised for low-cost inference to a 500-billion-parameter model designed for deep reasoning and long-horizon planning. This spectrum allows developers to deploy the right model for each agent within a workflow, balancing efficiency with intelligence.
Enterprise adoption of Nemotron is already gaining momentum. Industry leaders such as Accenture, Deloitte, EY, Oracle Cloud Infrastructure, Palantir Technologies, ServiceNow, Siemens, Synopsys Inc, and Zoom are integrating Nemotron models into real-world applications spanning consulting, engineering, enterprise software, and collaboration platforms.
AI search company Perplexity is also leveraging Nemotron within its agent routing system. CEO Aravind Srinivas said the platform uses Nemotron to optimise performance and cost across tasks. “We can direct workloads to fine-tuned open models like Nemotron 3 Ultra or use proprietary models when tasks require it,” he noted.
With Nemotron 3, NVIDIA is positioning open, agent-ready models as a foundation for the next generation of enterprise AI—where coordination, efficiency, and transparency matter as much as raw model scale.
