3 min readNew DelhiDec 20, 2025 12:27 AM IST
First published on: Dec 19, 2025 at 05:10 PM IST
In a move to bolster its open-source AI offerings, Nvidia has released its new and most “efficient family of open models” called Nemotron 3 this week. These models are designed to build and deploy accurate AI agents, as per the US-based chipmaking giant.
They come in three sizes, namely: Nano ( a 30 billion parameter-model for targeted tasks), Super (a 100 billion parameter-model for multi-AI agent applications), and Ultra (a 500 billion-parameter model for more complex tasks). The parameters of an AI model denote its size and generally serve as an indicator of its processing capabilities.
Based on benchmarks scores shared by Nvidia, the new Nemotron 3 Nano is more accurate than Alibaba’s equivalent-sized Qwen model and delivers better inference throughput than OpenAI’s GPT-OSS-20B. However, most notable is the fact that Nvidia has publicly released the data used to train its AI models along with the weights, training recipe, and framework.
The trillion-dollar company has made a fortune supplying Graphics Processing Units (GPUs) to its customers who are building AI models. However, with Nemotron 3, Nvidia appears to be taking a bigger step toward becoming an AI model maker itself.
It comes at a time when some of Nvidia’s biggest customers like OpenAI, Google, Meta, and Anthropic are looking to develop their own chips and reduce reliance on Nvidia’s technology over time.
“Open innovation is the foundation of AI progress. 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,” Nvidia CEO Jensen Huang said in a statement.
Nvidia is also taking a more transparent, open-source approach to building AI models compared to its potential US rivals. Publicly releasing the training datasets of Nemotron could help external developers modify the AI models more easily. “We believe open source is the foundation for AI innovation, continuing to accelerate the global economy,” Kari Ann Briski, vice president of generative AI software for enterprise at Nvidia, was quoted as saying by Wired.
Earlier this week, Nvidia also announced that it has acquired SchedMD, the company behind the popular open source workload management system called Slurm that is designed for high-performance computing and AI. In its announcement, Nvidia said that it will keep Slurm as an open-source and vendor-neutral software.
Alongside the Nemotron 3 models, Nvidia also released tools to help developers customise and fine-tune AI models. It has also launched libraries for users to train AI agents to autonomously handle tasks using techniques such as reinforcement learning.
Nvidia also released tools to help developers customise and fine-tune AI models. (Image: Nvidia) 
Share
