In a significant strategic shift, the American chipmaking titan Nvidia has unveiled its latest and most powerful suite of open-source artificial intelligence models, dubbed Nemotron 3. Announced on December 19, 2025, this move positions Nvidia not just as a hardware supplier but as a direct player in the AI model development arena, challenging some of its largest US-based customers.
Nemotron 3: A Trio of Open-Source Power
The new Nemotron 3 family is described by Nvidia as its most "efficient family of open models" designed specifically for building and deploying highly accurate AI agents. The models are available in three distinct sizes to cater to varying computational needs.
The lineup includes the Nemotron 3 Nano, a 30-billion parameter model optimized for targeted, specific tasks. Next is the Nemotron 3 Super, a more robust 100-billion parameter model intended for complex multi-agent AI applications. At the top sits the Nemotron 3 Ultra, a massive 500-billion parameter model engineered to handle the most intricate and demanding AI challenges.
In a notable benchmark, Nvidia claims the Nemotron 3 Nano outperforms Alibaba's similarly sized Qwen model in accuracy and surpasses OpenAI's GPT-OSS-20B in inference throughput, highlighting its efficiency.
Transparency as a Strategy
Perhaps the most groundbreaking aspect of this release is Nvidia's commitment to radical transparency. Unlike many of its rivals, the company has publicly released the complete training data, model weights, training recipe, and framework for the Nemotron 3 series. This open-source approach allows external developers worldwide to inspect, modify, and fine-tune the models with unprecedented ease.
"Open innovation is the foundation of AI progress," stated Nvidia CEO Jensen Huang. "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."
Kari Ann Briski, Nvidia's vice president of generative AI software for enterprise, reinforced this vision, telling Wired, "We believe open source is the foundation for AI innovation, continuing to accelerate the global economy."
Navigating a Shifting Competitive Landscape
This foray into open-source model development comes at a pivotal time for Nvidia. The company, valued at over a trillion dollars, has built its fortune by supplying critical Graphics Processing Units (GPUs) to AI giants like OpenAI, Google, Meta, and Anthropic. However, these very customers are now actively developing their own AI chips to reduce long-term dependence on Nvidia's technology.
By releasing powerful, transparent AI models, Nvidia is strategically diversifying its portfolio and embedding itself deeper into the AI software stack. This positions it as both a partner and a potential competitor to the firms that drive its hardware sales.
In a related move underscoring its commitment to open ecosystems, Nvidia also announced this week the acquisition of SchedMD, the company behind the popular open-source workload manager Slurm, used extensively in high-performance computing and AI. Nvidia has pledged to keep Slurm open-source and vendor-neutral.
To support developers, Nvidia released a suite of tools alongside Nemotron 3 for customizing and fine-tuning AI models. It also launched new libraries to help users train AI agents to autonomously handle tasks using advanced techniques like reinforcement learning, further cementing its role as an end-to-end AI platform provider.
