Nvidia has unveiled its first open-source AI model and is preparing several more

Nvidia has released a new artificial intelligence model, Nemotron 3.5 Lightning—its first open-source model. The release came just a couple of weeks after company CEO Jensen Huang urged the U.S. government to “avoid premature restrictions” that could drive innovation overseas, according to CNBC.

Nemotron 3.5 Lightning was developed primarily for AI agents—programs capable of autonomously performing tasks in the background, according to the developer’s press release. Nvidia claims that Nemotron 3.5 can run on a single graphics processing unit (GPU) in a standard PC. Users will be able to download, use, and modify the new version of Nemotron for free without Nvidia’s permission, the press release states.

The company has also released NeMo Switchyard software, which helps identify the most cost-effective and suitable AI model for a specific task.

Nvidia shares rose 0.5% to $218.5 during Tuesday's trading session. The previous day, they had fallen nearly 3% following the company's announcement that it was creating platforms for investing in AI in partnership with the world's largest financial firms.

For Nvidia, developing open-source AI is beneficial in terms of chip sales: such models must still run on graphics processing units (GPUs), and the lower cost of using them could stimulate demand compared to proprietary models from companies such as OpenAI and Anthropic, notes CNBC.

“Free AI should be great news for hardware manufacturers,” Huang said in an interview with Axios in July. “Free AI should be great news for chips.”

According to The Information, Nvidia is already developing a new family of artificial intelligence models called Nemotron 4, with the goal of competing with the world’s best open-source AI systems. According to the publication’s sources, Nemotron 4 is expected to have at least 1 trillion parameters, compared to the 30 billion parameters of the previously released Nemotron 3.5 Lightning. Nvidia has not yet completed the final training of the model, but company employees who spoke with The Information stated that it could be ready as early as late fall.

In July, Nvidia CEO Jensen Huang spoke out in defense of open-source artificial intelligence models and joined the debate that had erupted across the industry. Following the announcement of Kimi K3—a model developed by the Chinese company Moonshot AI that has narrowed the gap with the most powerful American models—there has been debate in Washington over closing the source code to prevent leaks. Lawmakers have expressed concerns that Kimi K3 could pose a threat to national security and that it may have been developed using others’ intellectual property through a method called distillation, according to CNBC. This method involves using the responses from a more advanced AI model to train a lighter model. For example, Nvidia used distillation to give Nemotron 3.5 Lightning capabilities comparable to those of larger models in the Nemotron family.

U.S. authorities have begun discussing whether to impose sanctions or other restrictions on Chinese model developers, similar to the restrictions on chip sales. Huang published an open letter in which he argued that “open-weight” models give companies greater control over their future, foster competition, and help lower prices. “Open models enhance security and cybersecurity, accelerate innovation and the spread of technology, and ensure technological sovereignty,” Huang wrote on the social media platform X.

A few days later, Nvidia, together with a number of companies—including Microsoft— launched an AI security consortium that will focus on the use of open models and open-source software in the field of cybersecurity.

Open-source AI has become one of the most talked-about topics in Silicon Valley. On August 10, Meta CEO Mark Zuckerberg published a detailed manifesto in support of open-source AI, coinciding with the company’s release of the Muse Spark programming model. “Our goal should be to make American open-source models the best in the world,” Zuckerberg wrote.

This article was AI-translated and verified by a human editor