What is Open-Weights AI?

Silicon Valley has been embroiled in a bitter battle over how artificial intelligence software should be built. Some have said that AI models should be “closed” because they are too powerful and dangerous to be open. Others said that AI models need to be “open sourced” for people to further develop the technology and build new businesses.

Amidst the debate, technologists battled over how much openness was enough, citing “open scales.” What are open scales? We will explain them.

First, what is open-source software?

When software developers decide to make their technology open source, they publish all the underlying computer code. They also release the code under an “open license” that allows other developers to use it without paying fees or royalties.

Publishing your code has several advantages.

One of them is security. Private code can be reviewed by company employees for flaws, but public code can be reviewed by anyone, so many more technologists remove bugs that could lead to cybersecurity incidents.

Another is faster innovation. Many companies have turned to open source software to introduce new products faster, instead of writing code from scratch. Today, open-source software underpins much of the modern Internet.

What are “open scales” in AI?

Artificial intelligence models learn their skills by pinpointing patterns in vast amounts of digital data. By finding patterns in books, newspaper articles, and other online documents, an AI model can learn to generate English on its own. The model stores these patterns as numerical calculations called “weights”. These values ​​determine how the AI ​​model “thinks” when responding to user prompts.

When technologists refer to open scales, they mean that the calculations or rules governing the AI ​​system are public. People who use the system can tweak these rules to suit their own needs — perhaps focusing the AI ​​system on a particular area, such as medical research or cybersecurity, or prioritizing certain information over all others. If the scales are not public, people must use the AI ​​system as designed by its creators.

Open scales don’t go so far as to reveal all the code that makes the system work. Some AI companies have released their core code, while others have only released the scales. And other companies have not released any details about the inner workings of their AI models.

Why do we care now?

This boils down to the AI ​​race and who has power over the technology.

Anthropic and OpenAI, leading artificial intelligence labs, have spent billions of dollars developing cutting-edge AI models that are mostly closed, meaning the companies don’t share anything about how those models were created. These models have been widely adopted around the world.

However, this concentration of AI power has raised concerns among tech giants such as Nvidia, Microsoft and Meta, who are also competing in the race. Many of these companies argue that sharing open scale models could allow the creation of other businesses on the cutting edge of the technology to expand the field.

There is also a security component. Some companies say that releasing open scales models allows for broader scrutiny, which can help companies fix vulnerabilities before those bugs are exploited.

Anthropic and OpenAI say that because some of the technology is so advanced, giving away all the code freely could be more dangerous, especially if the models are used by people with bad intentions. (It’s also very bad for their business.)

Unloading the weight of the AI ​​model is something of a compromise between open source and closed source software. Die-hard proponents of open source software argue that open scale models still do not go far enough in sharing information.

(The New York Times has sued OpenAI and Microsoft, alleging copyright infringement of news content related to AI systems. Both companies have denied the claims.)

How much of a factor is China?

China is a major factor in this debate.

China’s tech industry has embraced open source as a tool to catch up with AI development. Last year, Chinese startup DeepSeek shocked the tech industry when it released an artificial intelligence model that performed as well as the world’s best and published much of the underlying code as open source.

Since then, China has become the source of most of the world’s leading open-source models, including companies such as Alibaba, Z.ai and Moonshot AI. Also this month, Chinese leader Xi Jinping gave a speech portraying the country as a global champion of open access to AI development.

China’s strength in open-source artificial intelligence appears to pose a threat to America’s competitiveness in the technology, fueling a debate in Silicon Valley about how to move forward. Washington has also seen a back-and-forth as officials debate how or whether to regulate AI