27 Million Puzzles: What We Discovered About Difficulty in CrossEights

November 12, 2026

CrossEights has a very large collection of puzzles.

Nearly 28 million, in fact.

Each is a valid arrangement of eight eight-letter words—four across and four down—intersecting at sixteen points in an 8×8 grid.

That's a lot of possible games.

But here's the surprising part: when we evaluated those puzzles using our new difficulty classification system, only about 6.7% qualified for one of our four difficulty levels.

More than 26 million puzzles didn't make the cut.

What happened?

First, a little background

We recently redesigned how CrossEights classifies puzzle difficulty.

Our original approach relied heavily on word familiarity: puzzles with common words were considered easier, while those containing less familiar words were considered harder.

But after studying actual gameplay, we discovered that word familiarity wasn't enough to predict how difficult a puzzle would feel.

Two additional factors mattered enormously: how much useful information players could uncover with their opening hints, and how the words interacted at their intersections.

Our new difficulty model, Version 2.4, evaluates all three factors.

That gave us a much more precise way to identify puzzles that fit the experience we're trying to create at each difficulty level.

Then came the interesting question: What would happen if we applied this model to every puzzle in our library?

27,901,553 puzzles. Four difficulty levels.

We evaluated all 27,901,553 valid puzzle solutions against the new model.

Here's what we found:

Only 6.71% of our valid puzzle solutions satisfied the new classification requirements.

That doesn't mean the other 93% are broken puzzles. They contain valid words arranged in valid CrossEights grids.

But they don't meet the combination of vocabulary, opening-hint, and structural criteria we've established for our four difficulty levels.

We've become much more selective about the puzzles we serve.

And the distribution of those qualifying puzzles tells an interesting story.

Finding Easy puzzles is surprisingly difficult

Of the nearly 28 million valid grids, just 36,682 qualify as Easy.

That's approximately one in every 760 puzzles.

Why so few? And why do we care so much about getting Easy right?

CrossEights is an unusual kind of word puzzle. Most people encountering it for the first time have never played anything quite like it. There are no clues, just eight interlocking words, a handful of starting letters, and the opportunity to reveal more letters through hints.

Even experienced crossword and word-game players have to learn a new way of solving.

That's why having genuinely Easy puzzles is so important.

A new player needs a chance to discover how the game works: how to recognize words from partial letter patterns, how solving one word helps reveal others, and how to use hints strategically. Those early successes build the understanding and confidence that make more challenging puzzles enjoyable later.

An Easy puzzle that turns out to be unexpectedly difficult can interrupt that learning process. Instead of discovering the satisfaction of solving CrossEights, a new player may simply conclude that the game is confusing or frustrating.

Our Easy criteria are therefore deliberately demanding.

All eight words must come from familiar vocabulary. Opening hints must reveal a generous amount of useful information, distributed across the grid so that players have several opportunities to recognize words and begin making connections.

And the intersections must support a solving experience in which progress on one word helps unlock others.

A puzzle containing eight familiar words isn't automatically Easy. If those words intersect awkwardly or reveal unhelpful starting patterns, it may be a poor introduction to the game.

An Easy puzzle should teach you how to play CrossEights without feeling like a tutorial.

Getting that experience right matters well beyond the first puzzle. It's how new players develop the pattern-recognition and deduction skills that make Medium, Difficult, and Special Blend rewarding.

The surprising scarcity of qualifying Easy puzzles shows just how selective we need to be to create that experience.

Medium is the giant in the room

Of the 1.87 million puzzles that qualified for one of our difficulty levels, approximately 96% landed in Medium.

That's 1,799,782 puzzles.

The other three tiers combined account for fewer than 74,000.

This enormous concentration partly reflects how the model defines the different categories.

Medium accommodates a broader range of vocabulary and grid structures, while Easy and Difficult must satisfy more restrictive conditions.

But it also illustrates something important about puzzle difficulty: there's a vast range of possible experiences between an approachable, quick solve and a deliberately demanding one.

Medium isn't a single level of challenge so much as a family of puzzles that satisfy our criteria for a balanced solving experience.

That gives us a deep inventory from which to select engaging puzzles while maintaining a consistent overall character.

It also raises an interesting question for future development: can we learn more about the different kinds of challenge that exist within the Medium category?

For now, our goal is simpler—to make sure that the puzzles selected for Medium feel like they belong there.

Difficult puzzles are almost as rare as Easy ones

Only 35,403 puzzles qualified for Difficult.

That's remarkably close to the 36,682 that qualified for Easy.

But the reasons for qualifying are quite different.

Where Easy requires accessible vocabulary and generous starting patterns, Difficult is designed around more demanding word recognition, limited opening information, and greater resistance at intersections.

The objective isn't to make puzzles impossible, or to fill them with obscure words no one has heard of.

It's to create a solving experience that demands more careful deduction.

A good Difficult puzzle should eventually yield to a thoughtful approach. The satisfaction comes from figuring out how the pieces fit together, not from discovering that a word was missing from your vocabulary.

Our new criteria are intended to favor puzzles that create that kind of challenge.

And then there's Special Blend

Special Blend is the smallest category by an enormous margin.

Just 1,050 puzzles qualified—roughly four out of every 100,000 valid solutions.

That's a tiny fraction of the library.

Special Blend has a different purpose from our other tiers.

Rather than representing another step up a conventional difficulty ladder, it's designed to offer an unusual solving experience.

These puzzles feature uncommon letter combinations, distinctive consonant patterns, and structures that don't behave quite like ordinary grids.

Letters such as J, Q, X, Z, K, V, and W play a more prominent role.

We've also been careful not to confuse unusual patterns with obscure vocabulary. The aim is to create puzzles that make experienced solvers look at the grid differently, not to turn CrossEights into a test of dictionary trivia.

The very small number of qualifying Special Blend puzzles reflects just how particular we've become about what belongs in this category.

Why not simply use all 28 million?

It's a reasonable question.

After all, if a puzzle contains eight valid words arranged correctly in a grid, why not let people play it?

Because a mathematically valid puzzle isn't necessarily a good fit for the experience we want to deliver.

Consider two puzzles built from words of similar familiarity.

One offers useful opening hints and intersections that gradually reveal more information as you solve. The other offers scattered letters and awkward crossings that make it difficult to establish momentum.

Both are legitimate puzzles.

But they may produce very different experiences, even when assigned the same difficulty based on their vocabulary.

Our new model is an attempt to distinguish between those experiences before a puzzle ever reaches a player.

It's a form of curation, applied systematically across a very large collection of possible games.

Of course, no algorithm can perfectly predict how an individual will experience a particular puzzle. Vocabulary, solving habits, insight, and a little luck all matter.

But we can do a much better job of choosing puzzles whose structures match the challenges we're trying to provide.

A large library is useful. A well-classified library is better.

One of the advantages of having millions of possible CrossEights puzzles is that we can afford to be selective.

We don't need to use every valid grid.

Instead, we can apply increasingly precise criteria, drawing from a substantial collection of puzzles that meet our standards for each difficulty level.

That matters for both our Daily puzzle and Rival Mode, where players can choose their preferred challenge and play repeatedly.

It also matters for the competitive side of CrossEights. When scores reflect solving speed, hints, and deduction, a more dependable difficulty system helps make those comparisons meaningful.

The game will always contain surprises. We wouldn't want to eliminate them.

But we'd prefer those surprises to come from moments of clever recognition and unexpected deduction, rather than a difficulty label that doesn't match the puzzle.

The next 27 million puzzles

There's something satisfying about the result.

We began with nearly 28 million valid solutions and ended with a much smaller—but still enormous—collection of puzzles that fit our new difficulty classifications.

The most interesting lesson isn't that 93% didn't qualify.

It's that the difference between a valid puzzle and a well-calibrated puzzle is much more complicated than we'd originally appreciated.

Eight words, sixteen intersections, and a handful of opening letters create a remarkable range of possible solving experiences.

Understanding that range has given us a better way to select puzzles—and a much deeper appreciation for the surprisingly complex little game we've built.

We're looking forward to sharing the results with players beginning November 16.

Want to know why we redesigned the difficulty model in the first place? Read Why Easy Puzzles Weren't Always Easy: Rethinking Difficulty in CrossEights.