It says 300 Gh/s for NTLM and 200 kh/s for bcrypt.

There are 62^12≈10^21 12 character mixed case alphanumeric passwords. At 3*10^11 hashes/second, this would take ~3*10^9 seconds (~100 years) to brute force. For bcrypt, an 8 character password is fine (it would take ~30 years to brute force). Of course, you have no idea what hash function a breached site may have used, so best to err on the side of caution.

That said, avoiding password reuse is far more important than the strength of an individual password (not that it’s unimportant). The vast majority of breaches occur due to credential stuffing attacks on other logins, not from the initial breach itself.

The math says 12+ if randomly generated by something like a password manager and including uppercase, lowercase, numbers, and symbols. 95^12 is too big for pretty much anyone to crack for a very long time, if ever. There is a small caveat to this regarding what type of hashing is being used though. In nearly all cases, even in the case of relatively weak hashes, 12+ is enough, but there are hashes which truncate or break up inputs. Windows LanMan(LM) is probably the most common offender still seen these days that does this and any LM will always be breakable up to 14 characters due to its underlying algorithm leading to a weak separation of 7 + 7 characters. In the case of Windows hashing a default length of 15+ can help to avoid LM being stored but for legacy systems it may be unavoidable.

if you're using all valid characters (numbers, punctuation, upper and lowercase, etc) then every extra character grows the set of possible passwords exponentially, not linearly. 24 characters is a huge problem space, i don't know exactly how many possible characters there are but let's say it's 50, so that's 50^24 which is 41 digit number, a number so big I don't know what to call it and still in the realm of a password that would take so long to crack you would die before it was finished cracking.