The gap between those who can use these tools and those who cannot. It runs along the familiar lines of income, geography, language, age and disability, and there is a new dimension: paying for the better model is now a meaningful advantage, so capability is stratified by ability to pay.
Technical view
Access barriers include connectivity, device capability, subscription cost, and language coverage, since performance in lower-resource languages lags substantially and tokenisation inefficiency makes those languages more expensive per unit of text. Free tiers typically route to smaller models, creating a quality gradient. Counter-effects include falling inference costs and capable open-weight models running on modest hardware, which push the other way.
Related
Part of AI Terminology Index, section: Society and Impact.