Wordle Solver — Find Every Word Your Clues Still Allow

Quinvert has no connection to The New York Times Company or to the game it publishes. "Wordle" is that company's registered trademark and is used here only to say what this tool is for — the ordinary way one names the thing a tool works with. Nothing on this page is the game, reproduces the game, or is endorsed by anyone who makes it.

That matters for one practical reason beyond the legal one: the answers this page lists come from a public-domain dictionary, not from the game's own list of answers, and the section on the dictionary below explains exactly what difference that makes to what you see.

A green tile is the simple case: that letter, that square, no argument. A yellow tile says two things at once — the letter is in the answer somewhere, and it is not at the square you tried it at. Most people read only the first half, and a solver that does the same keeps offering words with the letter sitting exactly where the game just ruled it out.

A grey tile is the one that is almost always described wrongly. It does not mean "this letter is not in the answer". It means the answer has no more of this letter than the same guess already accounted for in green and yellow. If you guess a word with two Es and one comes back green while the other comes back grey, the answer has exactly one E — not zero. Reading that grey as a ban discards every correct answer and reports nothing, which is the most common failure in this class of tool and the thing this one is built around.

Guess GEESE against an answer of ERASE and the game marks one E present, one E correct, and the third E absent. The answer holds two Es; the guess spent three; the leftover has nothing to pair against and comes back grey. From those three tiles the true constraint is "exactly two Es" — a floor and a ceiling at once, derived by counting the tiles that came back coloured rather than by reading any single tile in isolation.

The same counting rule settles the mirror case. If two guesses each pin the same letter green at different squares, the answer needs two of that letter, even though neither guess said so on its own — each contributed one green, and only the merged board shows both. This solver reads the constraint off the whole board for that reason, and `test/wordle-tools.test.ts` holds it against four hundred answer-and-guess pairs rather than against a handful of examples.

Every remaining candidate is a possible answer, but they are not equally useful to guess. A word is good exactly insofar as it splits the field: words that would produce the same pattern of tiles are indistinguishable after you play it, so the best guess is the one that breaks the survivors into the most, smallest groups. This page scores each candidate by the average number of words that would still be standing after playing it — 1.0 means it finishes the puzzle outright — and puts the lowest first.

Suggestions are drawn from the words still possible rather than from the whole dictionary. A probe made of letters you want to test can sometimes learn more per turn, but it cannot win, and someone with three guesses left usually wants a word that could. Where the field is still larger than five hundred words — realistically only before your first clue — the split calculation is too slow to run while you wait, so the ranking falls back to positional letter frequency, which is the same question asked more cheaply.

Mistyping a tile is the most common way to get an empty result, and an empty result sends people off to doubt the dictionary rather than to check the input. So the marks are inspected for clues that cannot all be true before any word is looked at: the same square marked green as two different letters, a letter marked both correct and not-there at one square, a letter required more times than it is allowed, or floors that add up to more letters than the word has room for.

Where one of those turns up, the page says which square and which letter, in those terms. That is a different message from "no word matches your clues", which is a real and useful answer when the clues are consistent but narrow — and telling the two apart is the whole point of separating them.

Candidates come from the ENABLE word list — 172,823 English words released into the public domain, of which 8,636 are five letters long. The game's own answers are drawn from a shorter curated list, chosen by editors to be words ordinary players know. That list is the publisher's and is not reproduced here.

The consequence, stated plainly: this page returns every five-letter English word your clues still allow, which is a superset of what the game could be thinking of. Some results will be words the game would never use as an answer, though they are real words and most of them are accepted as guesses. Nothing here filters by how common a word is, and where a result looks obscure, that is why.

The length control above the grid takes anything from four to eight letters. The solving logic does not care how long a word is — the constraints are per-square and per-letter either way — so the same grid handles the six-letter and seven-letter variants that have grown up around the original, and the four-letter ones for children.

Changing the length re-reads the dictionary for that length and clears the grid, because a guess typed for one length says nothing about another. The word counts differ sharply: 3,903 four-letter words against 23,109 seven-letter ones, which is why a longer variant takes more guesses to pin down even though it feels like it should give more away.

No dictionary loads when this page does. Pressing Solve fetches one plain-text file — the words of the length you picked, 52 KB for the five-letter list — and every solve after that reuses what is already in memory. The matching, the counting and the ranking all run in your own tab.

Which means no guess you type is uploaded, there is no queue, and there is no reason to limit how often you press the button. The network panel during a solve is the way to confirm that rather than a claim to take on trust: one request for a word list, and nothing carrying what you typed.