Password strength checker

What is a password strength checker?

It takes a password and estimates how many guesses an attacker would need. Most of them do this by counting characters and character classes, which is why they reward a capital letter and a symbol and tell you that P@$$w0rd1! is strong.

This one scores the way guessing actually works: it looks for the things a cracking rule set tries first — dictionary words, leetspeak substitutions, keyboard runs, years, sequences — and prices your password at what those cost. It then shows you what a character-counting meter would have claimed, side by side.

Your password never leaves this page. There is no request, no logging and no server involved: open your browser's network tab and watch, or disconnect from the internet entirely and keep using it. That is the only design a tool like this may have.

How to use the password strength checker

1. Type or paste a password. It is analysed as you type, on your own machine. Use the show and hide button if you are not alone.
2. Pick an attacker. Rate-limited login form, unthrottled login form, a stolen database hashed slowly, or one hashed fast. The answer moves by eleven orders of magnitude across those four, which is the point of offering them.
3. Read the pattern table. It lists what an attacker would recognise in your password and what each piece costs to guess. If your password is a word with decorations, this is where that becomes visible.

The rules most meters enforce were withdrawn by the standard that invented them

NIST Special Publication 800-63B is the document behind most corporate password policy. Its current revision is unambiguous, and worth quoting rather than paraphrasing. On character classes: verifiers “SHALL NOT impose other composition rules (e.g., requiring mixtures of different character types) for passwords”. On expiry: verifiers “SHALL NOT require subscribers to change passwords periodically”. Both of those are prohibitions, not suggestions, and both are the opposite of what a typical signup form does.

What it requires instead is length. A password used as a single factor “SHALL” be a minimum of 15 characters — not the 8 that almost every meter treats as the bar. Where a second factor is present the minimum drops to 8. It also requires checking the password against a list of known breached ones, which is the one requirement this tool deliberately does not implement, for reasons in the limitations below.

The arithmetic behind that preference is not close. Eight characters drawn from all 95 printable ASCII characters gives 6.63e15 possibilities — 52.6 bits. Fifteen characters of nothing but lowercase letters gives 1.68e21, or 70.5 bits. The password that fails every composition rule has a search space 252,820 times larger than the one that passes them all.

And the composition rules are worse than merely useless, because there is one obvious way to satisfy them. Take a common word, capitalise it, swap a few letters for lookalike symbols, and put a digit and a punctuation mark on the end. We ran that recipe over the 10,000 commonest English words: 99.2% of the results satisfy all four character classes. The entire recipe — every word, both capitalisations, leetspoken or not, every trailing digit or year, every symbol — spans 29 million possibilities, about 24.8 bits. A meter that counts characters scores P@$$w0rd1! at 65.7 bits. It is overstating the difficulty by a factor of two trillion.

One more thing worth saying, because it cuts against advice this site would otherwise repeat. The famous four-random-words passphrase is 51.7 bits when the words come from the 7,776-word list most guides recommend — very slightly less than those eight printable-ASCII characters. Four words was good advice when it was written and it is thin now. Five words is 64.6 bits, six is 77.5, and the gap widens fast. If you take one number from this page, make it five rather than four.

Honest limitations

The estimate is a ceiling on weakness, not a certificate of strength. If the tool finds a pattern, that proves your password cannot be stronger than the pattern — that direction is sound. If it finds nothing, it proves only that this tool did not recognise anything, which is a much weaker statement. Treat a bad score as reliable and a good score as the absence of evidence.

The vocabulary is the 5,000 commonest words from twelve public-domain books, and that is the wrong corpus in a specific and disclosable way: the words people actually put in passwords are not the words Victorian novelists used. “Password”, “admin”, “letmein” and “sunshine” are absent entirely, while “monkey” and “dragon” are present because they occur in the novels. A real attacker's list starts with the words this one is missing, so the true guess count for a common password is lower than what you see here.

The blocklist check the standard requires is not implemented, and that is a deliberate refusal rather than an omission. Comparing against known breached passwords needs either a download of hundreds of megabytes or a request to somebody else's server. There is a well-designed protocol for the second — you send a short prefix of a hash and never the password — but it is still a network request from a page that promises there are none. Do that check separately, with a tool that is honest about making the request.

Crack times are entirely determined by the attack rate, which is why this page makes you choose one instead of quietly assuming. The four rates offered are order-of-magnitude figures for the scenario named, not measurements of any particular system, and the real rate depends on the hash the site chose, the hardware, and whether the attacker has the database at all. A meter that prints one time without stating its rate is not telling you anything you can check.

The engine carries 48 assertions. The load-bearing one is the bound: several thousand passwords built from the recipe above must all score as weak, paired with a control confirming genuinely random passwords of the same length do not. Two real bugs were caught that way — leetspeak decoding is one-to-many and a single-valued table meant W1th1! never resolved to “with”, and the corpus gap above, which is now asserted so it cannot drift unnoticed.

Why is it free?

Pattern matching against a word list is work your own machine does in a millisecond. There is no server to pay for, which is also why there is no account and nothing to upload — the two facts are the same fact.

If you want a password rather than a verdict on one, the password generator on this site makes them. Its entropy figure is exact where this page's is a bound, and the difference is instructive: a generated password's strength is known by construction, while a remembered one's can only be guessed at from the outside.