Oh, I noticed. I've also complained how Gemini 2.0 Flash is 50% more expensive than Gemini 1.5 Flash for small requests.
Also I'm sure if Google wanted to price Gemini 2.5 Flash cheaper they could. The reason they won't is because there is almost zero competition at the <10 cents per million input token area. Google's answer to the 10 cents per million input token area is 2.5 Flash Lite which they say is equivalent to 2.0 Flash at the same cost. Might be a bit cheaper if you factor in automatic context caching.
Also the quadratic increase is valid but it's not as simple as the article states due to caching. And if it was a bit issue Google would impose tiered pricing like they do for Gemini 2.5 Pro.
And for what it's worth I've been playing around with Gemma E4B on together.ai. It takes 10x as long as Gemini 2.5 Flash Lite and it sucks at multilingual. But other than that it seems to produce acceptable results and is way cheaper.