This project has already landed improvements in 3.10, and some much bigger improvements in 3.11. This work for 3.12 is "just" a continuation of that excellent effort:
25% number is from pyperformance benchmark suite, which you can replicate. Whether pyperformance is representative benchmark suite is another question.
It rubs people the wrong way but I always call out blanket statements. Generally languages get faster with each version and theres a lot of numbers thrown around, it doesn't mean your apps will get anywhere near that boost.
If you're lucky that one loop that concats strings got a few ms shaven off while that ORM youre using continues to grind the whole thing down.
https://www.phoronix.com/review/python-311-benchmarks/4