The measured reference
Wordle statistics
Every figure recomputes automatically from the site's data pipelines as the archive grows. Measured through 7 August 2026.
This page is the site's running statistical picture of Wordle: every measure the data supports, recomputed automatically each time the archive advances, and every number placed next to what it means. The short version first. The site has recorded 1,876 dated puzzles since 19 June 2021, drawn from 1,855 distinct words. Replayed under one fixed solver benchmark, those puzzles average 3.51 guesses, and 94.2% of them finish on the third or fourth guess. Measured this way, Wordle's difficulty is strikingly narrow: nearly everything lives between "solved on guess three" and "solved on guess four," with thin tails on either side. The interesting statistics are in those tails, and in what separates them from the middle.
The most useful thing the measurements show is what difficulty is not. It is not obscurity. The letters that feel exotic - J, Q, X, Z - are rare in answers, and the puzzles that stretch the solver are mostly ordinary words that happen to sit in crowded neighborhoods: change one letter and you get another plausible answer, then another. When a game drags to five or six guesses, the culprit is usually one of these one-letter-apart families, where each guess can eliminate only a candidate or two. That is why the repeats-and-near-misses section below is not trivia; it maps exactly the terrain where games are won on the fourth guess or lost on the sixth. If you take one strategic lesson from this page, take that one: when your remaining candidates differ by a single letter, stop guessing candidates and spend a turn on a word that tests several of the differing letters at once.
The second lesson is that the answer list is edited, not sampled. Compared with the 14,855-word accepted-guess list, the working answer pool avoids plurals almost completely: 29.2% of accepted guesses end in S, against 1.7% of the answer pool. The pool leans toward common, singular, everyday words. This is the single largest measurable difference between the two lists, and it has a practical consequence - a final-S guess spends one of your five slots on a letter position the puzzle almost never uses.
Has the game been getting harder? Here is the answer measured across year averages of the same benchmark, stated at exactly the strength the data supports. The gap between the easiest and hardest year averages is 0.13 guesses, within the roughly 0.13-guess range that chance alone produces for year-sized samples of this benchmark. Treat year-to-year movement as noise, not a trend. Day-to-day variance swamps the yearly movement, which is why a rough week always feels like proof the editors turned a dial. The drift table below shows the actual numbers so you can check that feeling against the record.
Two kinds of evidence appear on this page, and they answer different questions. Solver-benchmark figures are reproducible: the same lists and the same replay produce the same numbers, so they are the right tool for comparing puzzles and years to each other. Crowd figures - the opt-in Observatory sample - describe humans, but a self-selected slice of them, published only above fixed sample floors. Treat the benchmark as the ruler and the crowd numbers as testimony: valuable, but weigh the sample before quoting either. Where a crowd figure is missing below, that is the sample-size discipline working as designed, not a gap in the page.
Everything here regenerates from the same pipelines that power the rest of the site - the recorded answer archive, the two word lists, the solver replay lines and the Observatory aggregates - so the figures update on their own as new puzzles complete; the measured-through date above tells you exactly how current they are. The methodology section at the bottom names every source, including the published research this page leans on for context. For the interactive versions of these measurements, see the puzzle difficulty explorer and the Observatory.
How hard past answers have been
One fixed solver benchmark, replayed against every recorded puzzle, gives each answer a comparable "guesses needed" score. These are solver-recommended lines under one policy - a consistent ruler, not perfect play and not a claim about players.
Mean of the solver-benchmark line lengths over all 1,876 benchmarked puzzles. Use it as the comparison ruler for the yearly figures below, not as a target for human play.
Share of recorded puzzles the benchmark solves on guess three or four. The narrow spread means single hard days move a weekly average more than any real trend does.
Puzzles where one reply to the fixed opener left a single standout candidate. Rare by construction - a reminder that a fast solve is mostly the puzzle, not the player.
Puzzles that pushed the benchmark to five or more. Nearly all sit in crowded one-letter-apart families - the pattern to watch for in your own games.
Solver pressure is a percentile within recorded puzzles: 100 means "hardest for the solver so far," 0 the easiest - it ranks puzzles against each other rather than grading players. The full ranked list, with the reasons behind each score, lives in the difficulty explorer; for the stories behind the worst offenders, read the hardest Wordle words guide.
What kind of words become answers
Two lists power everything here: the site's reconstructed accepted-guess list (14,855 words) and its working answer pool (2,360 words). Membership in the pool never predicts what the NYT will pick - but the pool's shape tells you which letters earn their keep.
Mostly plurals. The guess dictionary is full of them, which makes final-S words feel like reasonable guesses.
The pool all but excludes plural forms - the single largest divergence between the two lists. In an endgame, a candidate ending in S is usually the one to drop first.
| Final letter | Answer pool | Guess list | Gap |
|---|---|---|---|
| E | 18.2% | 11.7% | +6.5 |
| Y | 15.7% | 10.3% | +5.4 |
| T | 10.9% | 5.4% | +5.5 |
| R | 9.2% | 5.0% | +4.2 |
| L | 6.7% | 3.6% | +3.1 |
The final slot is the most heavily edited position in the game: the answer pool concentrates its endings on a few familiar letters, while the guess list spreads across plurals and rare inflections. Combined with the S figures above, this is why endgame candidates that end in an unusual letter deserve extra suspicion.
| Letter | Answer pool | Guess list | Gap |
|---|---|---|---|
| E | 45.4% | 42.8% | +2.6 |
| A | 39.4% | 42.1% | -2.7 |
| R | 36.0% | 29.9% | +6.1 |
| O | 29.2% | 30.7% | -1.5 |
| T | 28.8% | 22.9% | +5.9 |
| I | 28.2% | 28.0% | +0.2 |
| L | 27.8% | 23.4% | +4.4 |
| S | 26.8% | 44.0% | -17.2 |
Positive gaps mark letters the answer pool over-uses relative to the guess list - the vowels and common consonants that make a word feel ordinary. Openers built from high-gap letters probe the pool the answers actually come from, which is the measured reason common-letter openers keep winning the Opener Lab rankings.
Repeats and near misses
Distinct words that have appeared as the answer on more than one date, out of 1,855 answer words recorded. Seeing a word come up does not retire it - the shortest recorded gap between appearances is about 1.4 years.
Pairs of recorded answers differing in exactly one position, like SHADE and SHAKE. Each pair is a fork where one guess cannot separate two live candidates.
Share of recorded answer words with at least one other recorded answer a single letter away. Crowded neighborhoods are the norm, not the exception - plan your mid-game around them.
| Word | First used | Used again | Gap |
|---|---|---|---|
| FEIGN | 8 July 2021 | 7 August 2026 | 5.1 yrs |
| GRIPE | 23 December 2021 | 6 August 2026 | 4.6 yrs |
| FLUME | 26 August 2021 | 30 July 2026 | 4.9 yrs |
The largest one-letter-apart families
- _ATCH BATCHCATCHHATCHLATCHMATCHPATCHWATCH 7 answers
- SHA_E SHADESHAKESHAMESHAPESHARESHAVE 6 answers
- _IGHT EIGHTLIGHTMIGHTNIGHTRIGHTSIGHT 6 answers
- _OUND FOUNDHOUNDPOUNDROUNDSOUNDWOUND 6 answers
Every word above has already been an answer, so none of this spoils a puzzle - but each family is still live territory: its unused siblings remain plausible answers. When your candidates collapse into one of these patterns, a probe word covering several of the differing letters beats guessing the family members one by one. The second-guess tool exists for exactly that situation.
Is Wordle getting harder?
Same benchmark, same lists, grouped by year - so the yearly averages are directly comparable. "Hard days" counts puzzles in the top tenth of solver pressure across all recorded history.
| Year | Puzzles | Avg guesses | Hard days | |
|---|---|---|---|---|
| 2021* | 196 | 3.51 | 10.2% | |
| 2022 | 365 | 3.48 | 9.3% | |
| 2023 | 365 | 3.45 | 9.3% | |
| 2024 | 366 | 3.54 | 9.3% | |
| 2025 | 365 | 3.53 | 12.3% | |
| 2026* | 219 | 3.58 | 14.2% |
The gap between the easiest and hardest year averages is 0.13 guesses, within the roughly 0.13-guess range that chance alone produces for year-sized samples of this benchmark. Treat year-to-year movement as noise, not a trend. The bars share one scale, so what you see is how little the yearly averages move: the visible story of this table is stability. The "hard days" column drifts more than the average does, which fits how the game feels - difficulty arrives as occasional brutal days, not as a rising floor. Before quoting a trend from this table, note that the answer list is curated by editors, not sampled from anything: any drift here is a change in editorial taste, and could reverse next month.
What players report
Games players explicitly chose to share with the Observatory. A self-selected sample: contributors skew toward engaged players, so read every figure below as describing them, not all players.
Distinct puzzles with at least one contributed game. Per-puzzle rows publish only at 10+ games; smaller samples are held back rather than shown.
The solve rate publishes once the sample clears the 10-game floor in every outcome cell - a deliberate guard against reading noise as signal.
Publishes once enough games back it. The floor exists so a handful of early contributors cannot masquerade as "the average player."
Contribute a game or read the full method on the Observatory page; the per-puzzle aggregates are downloadable as CC-BY CSV.
Methodology and sources
Figures on this page recompute automatically from the site's data pipelines each time the archive advances; no number is hand-entered. The measured-through date at the top of the page is the last completed puzzle in the data. Scope labels matter: "recorded puzzles" means the site's dated past-answer archive, the only source this site uses for past-answer claims; the "working answer pool" and "reconstructed accepted-guess list" are the site's maintained word lists, and pool membership never predicts a future answer. Benchmark figures are solver-recommended lines under one fixed policy - reproducible rulers, not proofs of best play.
- The site's own simulation method - opener scoring, the benchmark replay policy, and every artifact's provenance - is documented on How the solver works, with downloadable datasets on the data page.
- Greenberg, Ronald I. Effective Wordle Heuristics (2024). arXiv:2408.11730. arxiv.org/abs/2408.11730. Shows that cheap partition-style heuristics get within a few hundredths of a guess of exhaustive search - the practical justification for this site using fast heuristic scoring rather than full search.
- Lokshtanov, Daniel and Subercaseaux, Bernardo. Wordle is NP-hard (2022). arXiv:2203.16713. arxiv.org/abs/2203.16713. Proves that guaranteeing a solve within a given number of guesses is NP-hard in general - the theoretical reason "just play perfectly" is not a strategy, and heuristics are the honest tool.
- Selby, Alex. The best strategies for Wordle. sonorouschocolate.com. Exhaustive-search results proven exact for the fixed lists analyzed: SALET averages 3.4212 guesses on the original lists, and TARSE 3.5526 on the current expanded lists. An MIT group (Bertsimas and Paskov, An Exact and Interpretable Solution to Wordle, 2022) independently confirmed the SALET figure. These exact numbers are the ceiling this page's 3.51-guess benchmark average should be read against: the benchmark is deliberately a fixed reproducible policy, not a claim of optimal play.
Common questions
- What is the average Wordle score, according to the statistics?
- A fixed solver benchmark replayed over all 1,876 recorded puzzles averages 3.51 guesses. The site's opt-in crowd average is published only once at least 10 contributed games back it; below that floor the figure is held back rather than shown on a thin sample.
- How many Wordle answers have there been?
- This site has recorded 1,876 dated puzzles from 19 June 2021 through 7 August 2026, using 1,855 distinct words; 21 words have appeared more than once.
- What is the statistically hardest Wordle word so far?
- By the site's solver-pressure measure, PUPPY (Jun 30, 2026) ranks hardest among recorded puzzles. That measure scores how much a consistent solver replay struggled, not how players felt on the day.
- Is Wordle getting harder over time?
- The gap between the easiest and hardest year averages is 0.13 guesses, within the roughly 0.13-guess range that chance alone produces for year-sized samples of this benchmark. Treat year-to-year movement as noise, not a trend.
- Where do these Wordle statistics come from?
- From the site's own data pipelines: its recorded past-answer archive, a reproducible solver benchmark documented on the how-the-solver-works page, and opt-in player reports. External context comes from published research: Greenberg's 2024 heuristics study, the Lokshtanov-Subercaseaux NP-hardness proof, and Alex Selby's exhaustive-search results. Figures recompute automatically as the archive grows.