Password Generator & Strength Checker

Random passwords: length x log2(alphabet), all the way up

The alphabet is chosen by the preset: letters, digits and punctuation is 86 characters, letters and digits is 62, letters alone is 52, lower case alone is 26. Every character comes from a uniform draw on crypto.getRandomValues, and every class you kept is guaranteed at least once, so a site that demands a digit is not let down by chance.

The bit count is therefore `length x log2(alphabet)`, and the page prints it: twenty characters of the widest alphabet is about 128 bits, twenty lower case letters is about 94. Twenty is also where this page stops pretending you need more - past that point the interesting question is not the password, it is whether you reuse it.

Lookalikes, and why the two alphabet sizes differ

The lookalike filter removes I, l, 1, O, 0 and o, which are the characters people mistype when retyping a password from a screenshot. It costs about two bits out of 128 and saves a support call, and the page shows both sizes so the cost is visible.

The generator’s alphabet is 86 characters (24 punctuation marks, because quotes, backslashes and spaces are excluded so a password can be pasted into a terminal). The strength meter uses the conventional class sizes every meter uses, which add up to 95 with punctuation. Both numbers are printed, and the meter’s figure for a generated password is the conventional one - slightly optimistic, in the honest direction, and explained on the page rather than buried in a constant.

Passphrases: 324 words, 8.34 bits each, added in front of you

A passphrase is easier to remember and easier to type than a random string of the same strength, which is why the mode exists. The bank is 324 concrete short words - animals, food, places, objects - because function words make phrases that are hard to remember and easy to guess. Each word is worth log2(324) = 8.34 bits, so six words is about 50 bits and eight is about 67, before the optional digits and random capitals.

The words are drawn with replacement, which is the diceware rule: a phrase may repeat a word, and forbidding repeats would quietly make the search space smaller than the number printed. The strength of a phrase is the number of words rather than their length, because a phrase is attacked with a wordlist.

Measuring: what the number means, and what it cannot see

The meter counts the character classes in what you typed, multiplies up to `length x log2(alphabet)` and divides the resulting space by three attacker speeds: a fast hash on leaked data (100 billion guesses a second), a slow hash like bcrypt or argon2 (ten thousand) and an online form with rate limiting (ten). Printing the rates is what separates a number from a claim.

It also checks the patterns a cracker tries first: the most-used passwords, three or more identical characters in a row, alphabet and digit runs, keyboard rows, years, digits alone, and anything under twelve characters. Those flaws are listed in the page’s own words, and they are allowed to pull the word beside the number down - a password that is one character repeated sixteen times is not called strong, whatever its bit count says.

What it cannot see: whether the password has appeared in a breach (that needs a lookup, and this page makes no network requests), who else knows it, whether you have used it on five other sites, and whether it came from a pattern nobody has added to a wordlist yet. It also cannot see shape: a phrase of dictionary words built from words outside this page’s bank scores like a fortress by the character formula, which is why the flaws list says so and why the label is capped at "Fair" for anything that reads like a phrase.