> They are regulationmaxxing instead of benchmaxxing, that's my problem with them.

To those who is in know (r/localllama, r/sillytavernai), is well aware that Mistral models - at least the small, <=24b ones - are the least censored, even less than Chinese.