I wonder if some day we might see Archive.org organizations preserving older models as operating costs go down

I was thinking that somebody (Me? No... Somebody.) should start a foundation to start making archive copies of models... not for usage, but because every model is a (somewhat flawed) snapshot of the state of global knowledge as of the time it was trained. In fifty years, you'll be able to talk to Gemini 2.5 flash and get answers from the perspective of 2025. It might be a valuable for historic and sociological research.