Nvidia has introduced Nemotron 3, a new family of open AI models designed to help companies develop AI agents.
Nvidia has introduced new open source AI models. “Open innovation is the foundation of progress in AI,” says CEO Jensen Huang. “With Nemotron, we give developers transparency and efficiency to build agentic systems at scale.”
According to Nvidia, the market is shifting from classic chatbots to AI systems that can reason, plan, and control tools. Nemotron 3 should support that transition with models that are more efficient, require less computing power, and are more scalable in complex environments.
Three models for different tasks
The Nemotron 3 family consists of three variants: Nano, Super, and Ultra. Nano is the smallest model with 30 billion parameters and focuses on targeted tasks where speed and efficiency are key. Nvidia says this model is four times faster than its predecessor, with similar performance.
Super (100 billion parameters with 10 billion active parameters) and Ultra (500 billion parameters with 50 billion active parameters) are intended for heavier applications, such as environments where multiple AI agents collaborate. Ultra serves as a kind of central brain for more complex workflows, such as research, analysis, and strategic decision-making.
Nano is already available. Super is expected in early 2026, while Ultra will be released during the course of next year.