One growing role, especially in mathematics, is that of a host for "overlay journals": https://www.insmi.cnrs.fr/en/cnrsinfo/epijournaux-en-mathema...

I really like the idea. In short: arXiv, HAL and similar sites host the papers without any peer review (short of perhaps stopping crank spam) or access control. They're freely available to anyone. Authors then submit arXiv IDs (or similar) to the reviewers of "overlay journals", which then review and accept or not. The overlay journal accepts a paper by just adding it to its list of accepted arXiv identifiers, and that's that.

This ensures accessibility for all, keeps peer review, yet takes a lot of the practical hurdles away from actually running a journal. A journal can now just be a group of people who give thumbs up or down to arXiv identifiers, and if that group's conclusion start having weight in the community then it's become an important journal. Maybe they give away their listings for free, maybe they charge to read the reviews – it's really up to them what the business model (if any) will be.

It's really nice.