“Don’t use ‘open weight’ and ‘open source’ interchangeably”: Percona CEO on why AI terminology matters
The New Stack Paul Sawers
Percona’s CEO says ‘open weight’ isn’t the same as ‘open source’. He says the mix-up could water down open source beyond AI.
Based on reporting by The New Stack, Paul Sawers — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
AI companies love the word open. Peter Farkas doesn’t love what that word is starting to mean.
Speaking at Open Source Summit Europe in Prague on Wednesday, the Percona CEO and co-creator of FerretDB argued for a hard line: don’t use “open weight” and “open source” as if they were the same thing. His complaint is simple enough. Model weights are useful, but they are not the full recipe. They don’t give you the source code, the training data, or the process used to build the model.
That matters because open-weight models are already a serious part of production AI. In August, they made up 56% of tokens processed through Vercel’s AI Gateway and 60% of US-originating token consumption on OpenRouter, with Chinese-developed models making up the majority. So this isn’t an abstract naming fight. It’s about the words developers and buyers use while real traffic is flowing through these systems.
Farkas says the industry should stop pretending that downloadable weights equal the freedoms that open source has traditionally promised. He’s not dismissing open weights. He says they’re great for running models in your own environment, testing ideas, and, if you understand the risks, even using them in production. But calling that “open source” crosses a line, especially when the label starts doing marketing work it shouldn’t.
The concern goes beyond AI, too. Farkas worries that if companies get away with looser language here, the meaning of open source software itself gets softer. He points to the risk of something being called Apache 2.0 even when it can’t be used in the European Union. Meanwhile, the Open Source Initiative is still arguing with itself and everyone else about what open source AI should require, especially around training data. Its first definition came out in 2024, but the debate is reopening, with a fellowship, a new fellow, and more community discussions on the way.
My take — AI-written commentary, not fact-checked reporting
This is one of those rare fights where the pedants are right. If “open source” starts meaning “we let you download some weights and please don’t ask further,” then the term becomes a shrug with branding. AI companies can keep the perks of openness; they just shouldn’t get to borrow the credibility of software freedom for free.
Read more about this at: The New Stack