Nvidia released Nemotron 3.5 Lightning, an open-source AI model the company describes as "lightweight" and capable of running on a single GPU on a laptop or desktop2. The release marks Nvidia's first open-source model since CEO Jensen Huang publicly advocated for the open-source approach in late July.
Nemotron 3.5 Lightning is free for companies to download, use, and modify without obtaining permission or paying Nvidia. The model was developed particularly for agents — AI programs that can run autonomously in the background. Nvidia said companies including CrowdStrike, CodeRabbit, and Harvey have tested and customized the model.
Nvidia representatives said the company used distillation to give Nemotron 3.5 Lightning capabilities similar to its larger Nemotron models. The model will be available on HuggingFace and Nvidia's website.
Alongside the model, Nvidia released software called NeMo Switchyard, which can determine the cheapest and most appropriate AI model for any given task.
Separately, Nvidia also released NVIDIA-NemotronLabs-VoiceChat-11B, described as its first open-source full-duplex speech model1.
The Nemotron 3.5 Lightning release came less than three weeks after Huang posted on X for the first time, defending open-source AI models and inserting himself into a debate playing out across the industry. In that post, Huang argued that open-weight models allow companies greater control over their future, spur competition, bring down pricing, strengthen safety and cybersecurity, accelerate innovation, and enable sovereignty. He also urged the U.S. government to support open models while "avoiding premature restrictions".
"Free AI should be great for hardware. Free AI should be great for chips," Huang said in an interview in July.
Days after Huang's public advocacy, Nvidia launched an AI safety consortium with companies including Microsoft, focused on using open models and open-source software for cybersecurity.
The release arrives amid a broader push by major technology companies toward open-weight AI. Meta CEO Mark Zuckerberg published a lengthy manifesto arguing for open-source AI, and Meta recently released Muse Glimmer, an open-weight, 30-billion-parameter model distilled from its Muse Spark for on-device agentic workflows ctx. Meta also released a coding model called Muse Spark.
ANALYSIS Nvidia's decision to release a lightweight, single-GPU model optimized for agentic workloads aligns its open-source strategy directly with its hardware business — more developers building and deploying agents on Nvidia GPUs expands the addressable market for its chips.