“Don’t use ‘open weight’ and ‘open source’ interchangeably”: Percona CEO on why AI terminology matters

The AI industry has enthusiastically embraced the language of open source, even as some of its most prominent “open” models offer developers only a portion of what that term has traditionally promised.

Speaking at Open Source Summit Europe in Prague on Wednesday, Peter Farkas, co-creator of the open source MongoDB alternative FerretDB and now CEO of open source database services company Percona, has a simple request: “Don’t use ‘open weight’ and ‘open source’ interchangeably,” he says.

“’Open source’ is only going to remain ‘open source’ as long as we preserve the actual meaning and the freedoms that are behind open source, and open weights are not providing that,” Farkas continues.

“’Open source’ is only going to remain ‘open source’ as long as we preserve the actual meaning and the freedoms.”

The distinction is particularly relevant as open-weight models become a major part of production AI. In August, they accounted for 56% of tokens processed through Vercel’s AI Gateway and 60% of US-originating token consumption on OpenRouter, with Chinese-developed models accounting for the majority

An AI model’s weights are basically the numerical parameters produced during training, encoding the patterns the model has learned. Making them available means developers can download and run a model on their own infrastructure, even when its creator has not released the ingredients or process used to produce it.

“You don’t have the source code, you don’t have the training data, you only have the output of these two,” Farkas says. “So why would we call this open source in the first place?”

James Landay, director at Stanford’s Institute for Human-Centered AI (HAI), has drawn a similar distinction, arguing in an article published by HAI in August that access to model weights alone falls short of what developers and researchers need from genuinely open source AI.

“Open weights answer ‘can I run this?’” Landay said. “Open source answers ‘can I trust this, improve it, and build the next thing on top of it?’”

For now, Landay argues, the major AI labs are largely answering the first question while falling well short of the second.

“Open weights answer ‘can I run this?’ Open source answers ‘can I trust this, improve it, and build the next thing on top of it?’”

If “open washing wins”

It’s worth noting that some companies are already going further than releasing the weights. Xiaomi, for example, recently launched its MiMo-V2.6 models, livestreaming nearly a week of reinforcement-learning training through a public dashboard. The company also released more than 7,000 reinforcement-learning task environments, along with its RL training code and technical documentation, offering more transparency into how the models were fine-tuned.

Whether that’s enough for MiMo-V2.6 to qualify as “open source AI” is a separate question. Xiaomi’s release does, however, illustrate the spectrum Farkas is talking about: “open” AI releases can expose vastly different amounts of what went into building a model.

Farkas points to China’s DeepSeek as an example of the confusion, with its models widely described as open source despite the company releasing model weights rather than everything required to reproduce the models from scratch.

However, the ability to download and run those models remains valuable in its own right, he stresses.

“Are open weights a bad thing? No, open weights are great.”

“Are open weights a bad thing? No, open weights are great,” Farkas says. “You can run your models in your own environment, you can experiment with them, and if you understand the risks, you can also use it in production. The problem is when open weights are positioned as, ‘hey, this is as good as open source‘.”

Without a commonly understood boundary, Farkas argues, “open washing wins.” And he worries that accepting a looser meaning of open source for AI could ultimately weaken the definition for software more broadly.

“Companies are going to get away with calling something Apache 2.0 that [for example] you can’t use in the European Union,” Farkas says. “Open source AI is one thing, but if we talk about open source, and we let this happen to the core definition itself, that is going to go way beyond AI.”

A definition still under discussion

The Open Source Initiative (OSI) published its first Open Source AI Definition in 2024, establishing criteria around the freedoms to use, study, modify and share AI systems. But the definition remains contested, particularly over what information about training data should be required for an AI system to qualify as open source.

Responding to Farkas from the audience, Duane O’Brien, who became the OSI’s new executive director in April, acknowledged the criticisms of its original definition and made clear that OSI is reopening the discussion.

“We did have a conversation two years ago — it was an important conversation, and two years is a long time in this space,” O’Brien says. “We are in the process of reopening and having another set of conversations.”

The OSI recently launched an Open Source AI Fellowship, appointing Gabriel Toscano as its first fellow to help build consensus around what constitutes open source AI. The two-year program will explore potential revisions to the definition, while OSI also plans a series of community discussions, or “open source salons,” over the next two years.

O’Brien encouraged Farkas and others to participate, while acknowledging the criticisms already levelled at the original definition.

“The criticisms that have been lodged against open source AI definition ‘one’ are valid,” O’Brien says. “We have to continue that conversation.”