Generate a hash

How to use it#

1. Type or paste the text you want to hash, or switch to File and drop one in.
2. Every algorithm is computed as you type or as soon as the file is read, so there is nothing to press.
3. To check a download, paste the checksum its page publishes. The Checksum verdict says whether it matches, and the digest that matches turns green.
4. Switch between hex and Base64 to reformat the same digests, then copy the one you need.

What a hash can and cannot do for you#

A hash function turns any amount of input into a fixed-length value, and the same input always produces the same output. Change one character and the digest changes completely, which is what makes it useful for checking that a file arrived intact or that a value has not been edited.

It only works in one direction. You cannot recover the text from the digest, because the function throws information away. What you can do is guess: hash a candidate and see whether it matches. For short or common inputs that guessing is cheap, which is why a hash on its own is not a way to store passwords.

Storing passwords needs a function built to be slow and salted per user, such as bcrypt, scrypt, or Argon2. The bcrypt generator makes and checks bcrypt hashes. A SHA-256 of a password is fast to compute, which means it is also fast for somebody else to compute several billion times against a leaked database.

Which of these are still safe#

MD5 and SHA-1 are both broken in the sense that matters: it is practical to construct two different inputs with the same digest. MD5 collisions can be produced in seconds, and a SHA-1 collision was demonstrated in 2017. Neither should be used where somebody might benefit from forging a match, such as a signature or a security token.

They remain perfectly reasonable as non-adversarial checksums. Verifying that a download was not corrupted in transit, or that a cache key still points at the same content, does not involve an attacker choosing the input. A great deal of existing software emits MD5 for exactly this reason, which is why the tool still offers it.

CRC32 is not a hash in this sense at all. It is a checksum: a four-byte value built to catch accidental damage, such as a flipped bit on a disk or a dropped byte on a network, and ZIP, PNG, and gzip files each store one to check their own data. Anyone who wants to can change a file and keep its CRC32, so it proves nothing when someone might tamper on purpose. Use it to match a value a ZIP tool or a file format gives you, and a SHA digest for anything that has to be trusted.

For anything new, SHA-256 is the sensible default. It belongs to SHA-2, the family SHA-384 and SHA-512 come from too, so a request for a SHA-2 hash usually means SHA-256. SHA-384 and SHA-512 produce longer digests and are no slower on 64-bit hardware, but a longer hash is not more secure in any way that will matter to most work.

SHA-3 is a separate design, standardised in 2015 so there would be a standard to fall back on if SHA-2 were ever broken. It has not been, so SHA-3 is mostly asked for by a specification that names it. SHA3-256 and SHA-256 give different digests of the same input, so the name matters when you check one.

Keccak-256 is SHA-3 as it was before the standard changed its padding. Ethereum had already adopted it and kept it, so it is what Solidity's keccak256 computes. It shares SHA3-256's length and design and still gives a different digest for every input, which is why the page lists them as separate rows.

Hex and Base64 are the same bytes#

A digest is a fixed number of bytes: 4 for CRC32, 16 for MD5, 32 for SHA-256, 64 for SHA-512. Hex writes each byte as two characters, so a SHA-256 becomes 64 characters. Base64 packs three bytes into four characters, which makes the same digest about a third shorter.

Checksums published alongside downloads are almost always hex. Base64 turns up in HTTP headers and in subresource integrity attributes, where length matters more than being able to read it. Switching the setting here reformats the digests already on screen rather than computing them again.