The smallest crossword that can hold a long answer

No committee ever defined the midi. It is simply the name solvers reach for when a mini finishes too soon and a full newspaper grid is more evening than they have. Ours are 11×11 and 13×13 — and what separates them from a mini isn't the extra minutes. It's that a grid this wide can finally carry a ten-letter word.

30 answers in a typical 11×11 — about 3 times what a 5×5 mini holds, and 41 once you step up to 13×13
3.8 entries of seven letters or more in an average 11×11 grid. A mini has never held one and never can: its longest possible answer is five letters
59 distinct grid shapes across the two sizes — 168 layouts once mirrors and quarter-turns are counted, every one proved fillable before it shipped

What a wider grid actually buys you

A 5×5 is one room. Every answer touches every other, the whole puzzle is visible in a single glance, and a five-letter word is as long as the geometry allows. Widen the grid to eleven and something changes that has nothing to do with there being more clues: entries appear that are long enough to span the thing. CORNFIELD laid across the middle of a grid is not just a harder clue than CAT — it is a structural beam. It touches answers in two different corners at once, and the moment it falls, letters land in places you were nowhere near.

That is why a midi solves in a different rhythm. You work a corner until it dries up, jump somewhere else, and the long entries quietly stitch the regions together behind you. The characteristic midi moment is the one where a half-filled row of scattered letters stops being noise and turns, all at once, into a word — and then hands you three more.

Same generator, same page, different puzzle

Two grids dealt by the engine behind this page, drawn at the same scale. On the midi, every answer of seven letters or more is tinted.

5×5 mini
10 answers · longest 5 letters

11×11 midi
32 answers · 4 of seven letters or more

DALMATIANTREASURESBILLOWSRANCHES

Those 4 tinted answers cross 16 others, so between them they touch 20 of this grid's 32 entries. The 5×5 holds 10 answers and cannot, by construction, contain a single word of six letters.

The three-letter squeeze

Here is the arithmetic nobody puts on the box. Black squares are what break a wide grid into answers a normal vocabulary can supply — and the more of them you place, the more short entries you create. In our 11×11 shapes, 35% of all answers are three letters long. But three-letter words are the thinnest shelf in the language: our database holds 525 of them against 2,075 five-letter answers.

So a midi asks for three-letter words about 5.9 times as often as the word list can comfortably supply them, and asks for the long ones far less often than it holds them. That mismatch is where crosswordese comes from in every crossword ever printed — the ESNEs and OLEOs that exist only because a grid needed three letters and the setter had run out of real ones.

What the grid asks for, and what the shelf holds

Answer lengths as a share of an 11×11 grid, then as a share of the 8,800-word database it is filled from.

An 11×11 grid's answers
3 letters 35%
4 letters 27%
5 letters 15%
6 letters 10%
7 letters 6%
8 letters 3%
9 letters 2%
10 letters 1%
The word database
3 letters 6%
4 letters 17%
5 letters 23%
6 letters 21%
7 letters 18%
8 letters 7%
9 letters 6%
10 letters 3%

The two highlighted bars are the whole problem: three-letter answers fill 35% of the grid and make up 6% of the pool. Every other length is in surplus.

We haven't solved that — nobody has. What we have done is refuse the usual escape route. Every one of those 525 three-letter answers is a word somebody actually says, and every one arrives with a clue written for it by hand, which is why a repeated short answer here is ICE or EGG rather than a Middle English serf. The price is that short answers do come round again within a long session. The long ones essentially never do.

Why half the letters are yours alone

Look closely at a midi grid and you'll notice something a mini never does: some letters belong to only one answer. In our 11×11 shapes, 51% of the letters sit in an Across answer and a Down answer at once; the rest are unchecked, confirmed by one clue only. A micro or a mini is 100% checked — every single letter double-confirmed.

That isn't a shortcut; it's the standard European grid, and it is the only honest way to build a wide crossword from a small, everyday vocabulary. Full cross-checking at this width demands a lexicon of tens of thousands of words, most of which nobody has met, because every letter has to work twice. Half-checking buys the space for CORNFIELD instead — and it changes how you solve.

In a mini
Every letter is a witness.

Guess wrong and a crossing contradicts you almost immediately. Checking your work is nearly automatic — the grid does it.
In a midi
About half the letters stand alone.

An unchecked letter is never confirmed by anything else, so a plausible-but-wrong answer can sit undisturbed. Read the whole entry back before you commit, and lean on Check when a region starts feeling shaky.

Why the ladder stops at thirteen

The obvious next rung is 15×15, the size of a weekday newspaper crossword, and our engine can build one. We don't, and the reason is in the same arithmetic as before. Our 15×15 shape library asks for three-letter answers 43% of the time against 35% at 11×11, and cross-checks only 45% of its letters against 51%. Going wider with a vocabulary this size doesn't make the puzzle grander — it makes it thinner and gluier.

Past thirteen squares, a grid stops being a bigger crossword and starts being a sparser one.

A newspaper solves this by owning a vocabulary of tens of thousands of words, a good share of which exist purely to hold grids together. We'd rather stop at the width our 8,800 hand-clued everyday answers can fill properly. If the word database grows enough to change that answer, the fifteen will be waiting.

The size ladder, measured from the shape libraries
Size	Answers	Letters	Longest	7+ letters	Cross-checked
Micro 4×4	8	15	4	none	100%
Mini 5×5	10	19	5	none	100%
Midi 11×11	30	89	10	3.8	51%
Midi 13×13	41	121	10	4.2	47%

How to solve a midi without getting lost

The habits that finish a mini — read the clue, know it or don't — start to fail at thirty answers. What works instead is closer to how you'd clear a desk: in passes.

Take a free pass first. Read every clue once and write only what you're certain of. Nothing else. In a grid this size, that single sweep usually seeds all four corners, and it costs you no thinking at all.
Then work a neighbourhood, not a number. Forget the clue order. Go to wherever two or three letters are already sitting and finish what's around them — those clues are half solved and the cheapest in the grid.
Attack the long entries once they're a third full. A nine-letter answer with three crossings in place is often guessable outright, and it is the only move that pays out letters in two regions at once.
When a corner stalls, leave it. The best hint for a stubborn region is a letter arriving from outside it. On a mini there's nowhere to go; here there always is.
Check before a region hardens. With half the letters unchecked, a wrong answer can survive a long time and poison everything around it. Check flags wrong letters in red without telling you the answers, so it costs nothing but pride.

Where your puzzle comes from

There is no stock of midis behind this page. Every press of New puzzle picks one of 168 grid layouts and fills it, in your own browser, from 8,854 hand-clued answers carrying 15,019 clues between them — the same database behind our micro and mini crosswords. Your browser remembers every grid it has dealt you: an identical puzzle never comes back, and grids that merely resemble a recent one are quietly rejected too.

The solved counter above the grid is your lifetime total, kept on your device, and it counts honest finishes only. Check and Reveal word are there when you want them, but a puzzle you reveal your way out of doesn't add to the number.

Not every shape survived the audit. The grid layouts were originally found by a machine that scored one thing: whether the shape could be filled at all. Before this page opened, every candidate was made to clear three bars instead — fill at least thirty-eight times out of forty against the live word list, never stall while doing it (generation runs on the same thread that draws the page, so a shape that takes three seconds is a frozen tab), and contain no run of more than four black squares, because a longer bar reads as a wall dividing the grid into two smaller puzzles. A third of the candidates failed one of the three and were thrown out, and npm run smoke:midi re-proves all three rules every time the word database changes.

The midi is the size that prints

This is the classic pencil-and-paper shape: one grid and all its clues on a single sheet, which is exactly how our packs lay it out. Open Download a puzzle pack under the grid for a print-ready PDF — up to ten fresh puzzles, a full page each, answer keys collected at the back where you can keep them out of reach, an optional name-and-date line, and a large-print setting. Anything you print joins your browser's repeat history, so it won't turn up online later.

If the midi isn't the size you wanted

Going down, the mini is the five-minute version and the micro is two. The daily crossword deals one shared grid a day if you'd rather build a streak than a stack. And if it's the clues you're tired of rather than the size, a fill-in hands you every answer and hides only where they go, while an anagram crossword keeps the clues and stops them making sense.