Nvidia Unveils Nemotron 3 Open Models to Power Agentic AI
Nvidia introduces the Nemotron 3 family, an open lineup of three AI models built to drive agentic systems. The release shifts focus toward multi-agent orchestration and efficient reasoning.
Nvidia launches the Nemotron 3 family, a new lineup of open models and data libraries designed specifically to power the next generation of agentic artificial intelligence. The family includes three sizes—Nano, Super, and Ultra—all built on a breakthrough hybrid latent mixture-of-experts architecture that compresses memory needs, optimizes compute, and delivers high intelligence relative to their size.
This release addresses the rapid industry shift away from single-model chatbots and toward orchestrated AI applications that automate multiple models for proactive, intelligent agents. The smallest model, Nemotron 3 Nano, features 30 billion total parameters with 3 billion active and delivers four times the throughput of its predecessor, while the Super and Ultra models provide mid-range and deep reasoning capabilities for complex multi-agent workflows.
With this launch, Nvidia positions itself directly against a growing field of open and semi-open reasoning models as industry competition moves beyond raw parameter counts to focus on orchestration, reliability, and agent-centric performance. The company states that this new architecture helps drive down costs while still providing the high-speed, reliable reasoning that developers need to build agentic systems at scale.