A mismatch is the most common MD5 problem, and it often comes from the input, not the algorithm. You run md5("hello") in a script, paste the same word into an MD5 Hash Generator, and the two results differ. Usually one side hashed an extra space or line break, because MD5 counts every byte of the input.

Paste text, get its 32-character hexadecimal MD5 hash.
The MD5 Hash Generator converts text of any length into a 128-bit digest, written as 32 hexadecimal characters. The same input always gives the same digest, and the length never changes. MD5 is specified in RFC 1321. It is still common for checksums and cache keys, but it does not belong in anything that has to resist an attacker.
These values were calculated for the exact inputs shown:
| Input | MD5 hash |
|---|---|
| hello | 5d41402abc4b2a76b9719d911017c592 |
| Hello | 8b1a9953c4611296a827abf8c47804d7 |
| hello followed by one space | f814893777bcc2295fff05f00e508da6 |
| hello followed by a line break | b1946ac92492d2347c6235b4d2611184 |
A capital letter or one invisible character changes the whole value. That is the avalanche effect, and the last row explains the mismatch from the start. In a Linux shell, echo hello | md5sum hashes the line break that echo adds, while printf hello | md5sum returns the first value.
Say a large installer finishes downloading on a weak connection, and the publisher lists an MD5 value next to the link. This tool hashes text, not files, so run a local command on the downloaded file:
# Linux
md5sum installer.zip
# macOS
md5 installer.zip
# Windows (Command Prompt)
certutil -hashfile installer.zip MD5
If your value differs from the published one, even by a single character, the file is damaged or incomplete, so download it again. If the values match, the file arrived intact, and that is all a match shows. A checksum listed on the same page as the download only guards against corruption, whatever the algorithm, because whoever can change the file can change the number beside it. MD5 adds a second weakness. Researchers have been building MD5 collisions since 2004, and RFC 6151 states that MD5 is no longer acceptable where collision resistance is required. When tampering matters, compare a SHA-256 value taken from a trusted source, or verify a signature.
| Use case | Better choice | Reason |
|---|---|---|
| Detecting accidental file corruption | MD5 or SHA-256 | MD5 is fast and enough here; SHA-256 costs little extra. |
| Cache keys, ETags, duplicate detection | MD5 | Acceptable when nobody gains from forging two items with the same hash. |
| Password storage | Argon2id, bcrypt or scrypt | These are deliberately slow. MD5 and SHA-256 are fast, so an attacker can test many guesses per second. |
| Digital signatures and certificates | SHA-256 or stronger | MD5 is ruled out, and SHA-1 has been phased out for signatures too. |
| Git object names | SHA-1 or SHA-256, set by Git | Git has used SHA-1 historically and also supports SHA-256 repositories. |
Some frameworks build ETag headers from a hash of the response body. To see which ETag a page actually returns, read its response headers with the Get HTTP Headers tool.
Not by running a formula backwards, because the hash does not contain the original text. Guessing does work, though. Short or common inputs such as password123 can be found by hashing candidates and comparing the results, and lookup tables of known hashes exist.
It is a pair of different inputs that produce the same MD5 hash. This is not only a rare accident: collisions can be built on purpose, and RFC 6151 notes that such attacks have been applied to X.509 certificates.
Not for storing passwords, since MD5 is too fast for that job. Keep real secrets out of any online tool as well, and hash them with a local command or your own code.
Yes. It is free for webmasters and developers, with a reCAPTCHA check before the hash is generated.