Not “free to use” — free to actually look inside, which is a stronger and rarer claim.
Eric Raymond’s 1998 essay “The Cathedral and the Bazaar” is the founding text of open-source development philosophy — not a peer-reviewed study, but the argument that gave the movement its name and its case: software developed in the open, with many independent contributors able to inspect and modify the code, tends to be more reliable than software built behind closed doors by a small controlled team, because bugs are shallow when enough eyes can see them.
Applied to AI, “open source” carries a sharper stake than it does for ordinary software. A closed model is only ever available as a black box behind an API — you send a prompt, you get an output, and you have no way to inspect what happened in between or run it without the provider’s servers, their pricing, and their data policy. An open-source model — the kind distributed through Ollama, or released by labs like DeepSeek and Qwen — can be downloaded, run offline, and inspected line by line. That’s the actual axis this essay is describing: not “cheap versus expensive,” but “audited and yours to run” versus “trusted and rented.”
Source: Raymond, E.S. (1998), “The Cathedral and the Bazaar,” First Monday, 3(3).