NVIDIA Supercharges Agentic AI with Faster, Cheaper Open-Source Models
In Focus
- NVIDIA’s latest AI models come in three sizes, Nemotron 3 Nano, Super and Ultra
- Nemotron 3 Nano is the smallest of the open-source AI models
- The new AI models offer high performance and greater transparency
NVIDIA has unveiled open-source AI models, which it says will be faster, smarter, and cheaper compared to the previous ones. News of NVIDIA’s open source AI came as the company released Nemotron 3 Nano, which is the smallest of the AI models, on December 15, 2025. According to Yahoo Finance, the AI models can write, code, and perform other tasks.
NVIDIA Open-Sources AI Models for Research
NVIDIA is popular for producing AI chips that companies like OpenAI use to train their AI models. Recently, the chipmaker developed chip verification technology that might show which country its semiconductors are located in a bid to curb smuggling.
Besides AI chips, the company also offers various AI models in different areas, including self-driving vehicles and simulations. Unlike OpenAI, NVIDIA’s open-source AI initiative makes its AI models accessible to researchers and other tech companies. For instance, Palantir Technologies has already started integrating the chipmaker’s AI model into its products.
NVIDIA’s AI developer ecosystem update involves the release of three sizes of Nemotron 3 AI models. These are Nemotron 3 Nano, Nemotron 3 super, and Nemotron 3 Ultra. Here are the main features to expect with each of these sizes:
- Nemotron 3 Nano is a 30B parameter model designed for highly efficient tasks
- Nemotron 3 Super is a 100B parameter model designed for multi-agent applications and offers high-accuracy reasoning
- Nemotron 3 Ultra is a 500B parameter model designed for large reasoning and complex applications
AI Models are Built on the Mixture-of-Experts Architecture
NVIDIA, which is now valued at over $5 trillion, used the hybrid mixture-of-experts architecture in the development of its Nemotron 3 models in order to boost efficiency and scalability. The updated NVIDIA AI platform will offer businesses high performance and greater transparency as they develop multi-agent autonomous systems.
“We believe that we are uniquely positioned to serve a wide range of developers who want full flexibility to customize models for building specialized AI by combining that new hybrid mixture of our mixture-of-experts architecture with a 1 million token context length,” NVIDIA Vice President for Generative AI Software Kari Briski told reporters as stated by Venture Beat.
Taking on Chinese Tech Firms
NVIDIA updates its open-source AI platform at a time when models from Chinese firms like DeepSeek, Alibaba Group, and Moonshot AI are gaining traction in the tech industry. Already, Airbnb has started using Alibaba’s open-source AI model known as Qwen.
Despite the competition, multiple companies have started adopting Nemotron 3 AI models. They include Palantir, CrowdStrike, Accenture, Deloitte, Oracle Cloud Infrastructure, Perplexity AI, Zoom, ServiceNow, and Siemens.
In a move that could benefit NVIDIA, Meta Platforms is reportedly weighing the option of closed-source models. Should the social media giant actualize this plan, NVIDIA could become one of the prominent open-source AI model providers in the U.S.
