Wordle Solver: Find Every Word That Still Fits Your Colours

By KingPin · Ukraine

You have five tiles of feedback and a shrinking number of guesses. The Wordle Solver takes exactly that state of the board and returns every word that survives it, so the next guess is a choice between two or three candidates instead of a shrug. It covers 61 languages, two word pools per language, and it runs the duplicate-letter maths the way the game does.

What Does a Wordle Solver Actually Compute?

A solver is a filter with a count on the end. It starts from a word list, removes everything your tiles rule out, and hands back what remains. No ranking, no guessing, no hidden heuristic scoring, because a filtered list is already the honest answer to the question you asked.

That matters when you are stuck. Take the classic ending where the pattern ends in -IGHT and the candidates are LIGHT, MIGHT, NIGHT, SIGHT and TIGHT. A ranking tool tells you which of those five it likes best, which is a coin flip with extra steps. A filter tells you that all five are still alive, so you stop believing the guess you already committed to in your head and go looking for a word that tests three of them at once.

Every marking you enter is a hard constraint. A green tile fixes a letter to a position. A yellow tile removes that position for the letter while keeping the letter in play. A gray tile removes a letter outright, unless the same letter is green or yellow somewhere else in the row, in which case it caps how many copies the answer can hold. The output is the set that satisfies all of those constraints together.

The count matters as much as the list. If the solver reports 221 candidates, your opener told you almost nothing and guess two should be another wide probe. If it reports two, stop guessing and start eliminating. Reading the number is how you decide which of those two moves to make.

How Do You Enter a Row?

Five tiles hold one letter each, and each tile cycles through its three states when you click it: gray first, then yellow, then green, back to gray. Type the word you played, then click the tiles until they match the colours on your board.

Two dropdowns sit above the grid. The first picks the language, 61 of them from Afrikaans to Chinese and Vietnamese. The second picks the word pool, either the curated answer list or every legal guess. Both choices load a different dictionary behind the tiles, and the page keeps them cached after the first fetch, so switching back is instant.

Press Find Words and the matching list appears as chips with a count above them. Press Clear to wipe the tiles and start the next row, which is useful when you want to test a hypothetical position rather than the board in front of you. The solver does not care which order you mark things in.

One detail worth knowing before you start: the input accepts letters from any alphabet, so Cyrillic, Greek, Hebrew, Arabic, Devanagari, Thai, Hangul and Japanese all pass through. Type the letter, mark its colour, and the filter runs on that language's file. There is no separate mode to switch into and no transliteration step.

Why Do Repeated Letters Trip Up Most Solvers?

Duplicates are where hand-rolled filters quietly break. Guess PAPER when the answer holds a single P, and Wordle colours the first P and leaves the second one gray. A filter that treats gray as "letter absent" now deletes P from its alphabet and throws away the answer. The tile never said the letter was missing. It said there is no second P.

The game scores a row in two passes. Greens are claimed first, one copy of the letter per green, walking the whole word before a single yellow is awarded. Then the remaining tiles are read and marked yellow only while unmatched copies of that letter are still available in the answer. Once the copies run out, every further tile of that letter goes gray.

The practical rule follows from that order: count the green and yellow copies of a letter first, and read a gray copy of the same letter as a cap on the count rather than a deletion. The Wordle Solver stores a minimum and, when a gray extra is present, an exact number.

This is not a corner case you can skip. In the English answer pool of 2,315 words, 749 of them, 32.4 percent, contain a repeated letter. Across the wider list of 12,972 legal guesses the share is 35.8 percent, and 172 answer words carry two Es. Any solver that deletes a letter because one copy of it came back gray will fail on roughly a third of the puzzles it meets.

Answers Pool or All Allowed Words: Which List Should You Pick?

The two pools answer different questions. The curated list is what the game draws its answers from, so words in it are plausible solutions and words outside it are not. The wide list is everything the game will accept as a guess, which is useful in the opposite direction: when you are hunting for a probe word that tests four letters at once, you want the full vocabulary, not just the answer pool.

Solving, use the answers pool. The candidate count stays small enough to read: 2,315 words for English instead of 12,972, 2,309 for Russian, 2,309 for German. The letters you already know constrain the pool hard, so the list you see is close to the real short list.

Probing, switch to the wide pool. It contains every string the game accepts, which is what you need when the answer pool has gone thin but you still have letters to test. The wide pool also tells you whether a word you are about to play is legal at all before you waste a turn on it.

A consequence of the split: a word can be absent from the curated list and still be accepted by the game. Filters that only load the answer pool will tell you there are no candidates left while your board still has a legal move. If you ever see an empty result you cannot explain, switch pools and look again before you assume the constraints contradict each other.

Which Languages Does It Cover?

The site ships 122 dictionaries across 61 languages, one pair per language, about 7.5 megabytes of word data in total. The English pair comes from the game's own published lists: 2,315 answers and 10,657 additional guesses, 12,972 words together. The other 60 pairs are built from frequency-ranked vocabulary for each language and cut down to five characters.

Some languages have thin pools, and the reason is structural rather than a bug: a five-character filter removes almost everything from a vocabulary whose common words are two or three characters long, so Hindi arrives with 22 words and Malayalam with 25. Those pools are still correct for what they contain, but a Chinese or Thai puzzle needs the wide list to stay useful, and a few of the smaller dictionaries are best treated as a hint source rather than a full solver.

A Worked Example: Ten Candidates in Three Markings

Start with English and the curated pool: 2,315 words on the board, nothing marked. Play SLATE and watch it come back with five gray tiles. All five letters leave the alphabet, and the filter drops the pool to 221 words. That is your opener doing its job, and 221 is still a large enough field that the next guess should be another probe rather than a commit.

Now suppose the second row shows an O, yellow, in the second tile. The letter is in the word but not there, and the pool falls from 221 to 62. The chips read bingo, brood, brook, broom, brown, buxom, and 56 others, which is the point where a human player starts guessing and a filter starts narrowing.

Add a green C in the first tile. The pool lands on 10 words: chock, choir, chord, crock, crony, crook and four more. Two more markings, either of the next two letters, will take it to one. That progression, 2,315 to 221 to 62 to 10, is a filter doing arithmetic, and every number is reproducible by hand from the same file.

The same logic runs on the other languages without translation. Take Russian at the wide pool of 18,648 words and mark Ч gray. The list falls to 17,642 words, which is what the interface reports when the query is run live on the page. A second gray letter cuts it again by roughly the same proportion, which is why two clean probes usually beat one clever guess.

Where Does the Solver Stop Helping?

Two situations stay hard no matter how good the filter is. The first is the coin flip at the end: two candidates differ by a single letter and you have one guess left. The solver will show both and refuse to choose, correctly, because the information is genuinely symmetric.

The second is the row you cannot mark honestly, usually because you mis-clicked a tile two turns ago and built a constraint set on a wrong colour. If the result looks impossible, clear the grid and re-enter the board from the top. It is faster than debugging a bad marking, and the solver has no way to know which of your tiles was the typo.

Where the tool earns its place is the middle of the puzzle, rows two through four, when the candidate list is too long to hold in your head and too short for another throwaway word. That is the window where a count and a list change the next move from instinct into arithmetic.

What Happens in the Browser?

The dictionaries are plain text files fetched from the site and cached by the browser after the first load. The filtering runs in the page. Nothing is uploaded, nothing is stored, and there is no account, no query limit and no login to keep alive between puzzles. Your board state stays on your device, which matters if you play the same puzzle as other people in a competition or a private group.

Nothing is guessed on your behalf either. The solver returns the words that fit the markings you entered and stops there. It will not reveal today's answer, and it cannot, because a filter works on evidence rather than predictions.

KingPin

Ukraine · SO reputation 2000 · Badges: 1🥇12🥈8🥉 · SO member since 2010