The Hardest Sudoku Ever Made, and What Makes a Puzzle Brutal

No single puzzle is officially "the hardest sudoku," because difficulty depends entirely on which method you measure it with — but a handful of grids have become famous for defeating both humans and solving software. The best known is a puzzle nicknamed "Everest," constructed by Finnish mathematician Arto Inkala, which has been widely reported as among the most difficult ever published.

What makes those puzzles hard is not the number of clues. It's the depth of reasoning required before any cell can be filled.

What Difficulty Actually Measures

Sudoku difficulty ratings are generated by running a solver against the grid and recording which techniques it needed. A puzzle solvable with scanning alone rates easy. One requiring naked pairs rates hard. One requiring chains of conditional logic rates extreme.

The brutal puzzles share one property: almost every deduction is deep. In a normal hard puzzle, an advanced technique cracks something open and then several easy placements follow. In a genuinely brutal puzzle, you apply an expensive technique, gain one cell, and are immediately back to needing another expensive technique. There's no downhill stretch.

The 17-Clue Floor

There's a hard mathematical limit worth knowing: no valid sudoku puzzle with a unique solution can have fewer than 17 starting clues. This was proven in 2012 by an exhaustive computational search that checked every possible 16-clue configuration and found none that produced a single unique answer.

That said, a 17-clue puzzle is not automatically the hardest kind. Many 17-clue grids are only moderately difficult, while some puzzles with 22 or more clues are far nastier. Clue count sets a floor on possibility, not on difficulty.

What Actually Makes a Grid Brutal

Clue placement, not clue count. Clues clustered in a few boxes leave large empty regions with weak constraints. Clues spread thinly across all nine boxes constrain everything a little, which sounds helpful but often means nothing is constrained enough to resolve.

No cheap entry point. Most puzzles have at least one obvious opening move. Brutal grids have none — the first placement itself may require an advanced pattern.

Long dependency chains. Some extreme puzzles require following a hypothesis several steps deep ("if this cell is a 4, then that one is a 7, which forces this one, which contradicts...") before a contradiction appears. Humans find this exhausting because it means holding a branching state in working memory.

Multiple simultaneous constraints. The hardest patterns require noticing relationships across three or more rows and columns at once, rather than within a single unit.

Famous Hard Puzzles

"Everest" (Arto Inkala). Published around 2012 and widely reported at the time as the hardest sudoku ever constructed. Inkala built it deliberately to maximize the depth of reasoning required, and it's still the puzzle most often cited in this conversation.

"Platinum Blonde." A grid that circulated widely in solver forums and scores extremely high on several automated difficulty metrics. It's a favorite benchmark for testing solving software.

"Golden Nugget" and "Discrepancy." Two more constructed grids that appear near the top of most computational difficulty rankings.

These aren't newspaper puzzles. They're deliberately engineered edge cases, closer to a stress test than to entertainment — and worth attempting mostly to see what the ceiling looks like.

Difficulty Tiers, Practically

Can You Actually Solve One?

Yes, with three things: complete and accurate pencil marks, comfort with advanced techniques, and patience for chain logic. What you cannot do is brute-force it by feel.

A practical approach if you want to attempt a famously hard grid:

1. Mark candidates twice — do a full pass, then verify it cell by cell. One error makes the rest impossible.
2. Expect no easy cells. Don't interpret the absence of an obvious move as a sign you've missed something simple.
3. Work in sittings. These puzzles reward fresh eyes far more than persistence.
4. Keep the grid recoverable. Use a pencil, or a digital tool with undo, because a single wrong placement forty moves in is otherwise unrecoverable.

Guessing still isn't required. Even the hardest constructed puzzles have a complete logical path. The path is just long, and finding it is the entire challenge.

The Techniques That Only Appear at the Top

Below extreme difficulty, most solvers never encounter these. At the very top they're unavoidable.

Forcing chains. Pick a cell with two candidates. Assume the first, follow the consequences as far as they go, and see whether you reach a contradiction. If you do, the other candidate is correct. If both branches lead to the same conclusion about some third cell, that conclusion holds regardless — which is the more useful version, because it produces a result without needing a contradiction.

Coloring. A visual form of chain logic. Mark cells where a candidate appears in only two places within a unit, alternating two colors along the chain of links. If two cells of the same color ever appear in the same unit, that color is false and every cell of the other color is true.

Unique rectangle. Uses the fact that a valid puzzle has exactly one solution. If four cells form a rectangle spanning two rows, two columns and two boxes, and all four hold the same two candidates, the puzzle would have two solutions — so that configuration cannot occur, and some candidate must be eliminated to prevent it.

ALS (almost locked sets). Advanced enough that most solvers never learn it. Two sets of cells that are one candidate short of being locked can be linked to produce eliminations neither would give alone.

The honest summary: forcing chains and coloring are learnable in an afternoon and will get you through most extreme puzzles. The rest are for people who find the puzzle itself less interesting than the theory behind it, which is a perfectly reasonable place to end up.

A More Useful Target Than "Hardest"

For most solvers, chasing the single hardest puzzle is less satisfying than steadily climbing tiers. The jump that actually improves your solving is expert to extreme, because that's where advanced patterns stop being optional. Generating a run of puzzles at one fixed level with a sudoku puzzle maker and staying there until it feels routine builds more real skill than one heroic attempt at Everest.

How Difficulty Ratings Are Actually Produced

Understanding where a rating comes from explains why two puzzles labeled "hard" can feel completely different.

Solver-based rating. The standard method. A program attempts the puzzle using techniques in order of complexity, records the hardest technique it needed, and assigns a rating accordingly. A puzzle solvable with singles alone rates easy; one requiring a swordfish rates extreme.

Weighted scoring. More sophisticated systems score every deduction the solve required and sum them, so a puzzle needing X-Wing once rates lower than one needing it four times. This is why some puzzles feel relentless while others have one hard moment and then open up.

Branching depth. For the very hardest grids, raters measure how deep a hypothetical chain must run before producing a contradiction. A puzzle requiring a seven-step chain is in genuinely different territory from one requiring three.

The practical consequence for a solver: ratings are not comparable across sources. One publisher's "expert" may be another's "hard." If a book or app feels miscalibrated against what you're used to, it probably is, and that's a property of the rating system rather than of your ability.

Why Constructed Puzzles Feel Different

The famous hard puzzles weren't found — they were engineered, and it shows in how they play.

A newspaper puzzle, even a difficult one, is generated and then rated. It has whatever character the generator happened to produce, which usually means some easy stretches mixed with a few hard moments.

A deliberately constructed hard puzzle is built backwards from the difficulty target. The constructor removes clues specifically to eliminate cheap deductions, tests the result against a solver, and keeps adjusting until every remaining path demands expensive reasoning. There is no easy stretch because the easy stretches were designed out.

That's why these puzzles feel airless. A normal hard puzzle gives you a small win every few minutes; a constructed extreme puzzle can go twenty minutes between placements. Knowing that in advance makes the experience considerably less demoralising — you aren't missing something obvious, because nothing obvious was left in.

What Makes a Puzzle Hard for a Human vs. a Computer

These diverge more than people expect, and it explains some odd ratings.

A computer finds chain logic trivial — it can follow a twelve-step implication chain instantly and without error. What costs a computer effort is search breadth, not depth.

A human finds the opposite. Following a chain three or four steps deep is genuinely taxing because it means holding a branching hypothetical state in working memory while continuing to read the grid. But a human spots visual patterns — an X-Wing sitting in plain view — almost instantly, where a program has to check systematically.

So a puzzle rated "extreme" by a solver program because it needs a long chain may feel merely hard to a person who spots a shortcut. And a puzzle rated moderate may feel brutal to a human if its key deduction is a pattern that's easy to describe and hard to see.

The practical upshot: trust your own experience over the label. If a puzzle marked "hard" is defeating you, it may genuinely be harder for humans than its rating suggests.

Keeping a Hard Puzzle Recoverable

Nothing is worse than being ninety minutes into an extreme puzzle, finding a contradiction, and having no way back.

Use pencil, and press lightly. Obvious and routinely ignored.

Mark givens distinctly. Circle them, or write your own placements in a different color. When you need to unwind, knowing which numbers were never yours is essential.

Take a photo at milestones. After the initial pencil-marking pass, and again whenever you've made a significant breakthrough. A phone photo is a free save point.

Write down your chain assumptions. If you're following a hypothetical, note on paper which cell you assumed and what value. Chains abandoned mid-way and half-remembered are a common source of corrupted grids.

If you find a contradiction, stop immediately. Don't keep placing numbers hoping it resolves. Go back to your last photo or, failing that, restart the pencil marks from the givens.

Diabolical, Evil, Impossible: What the Labels Mean

Above "expert," publisher naming stops being descriptive and starts being marketing. It helps to know roughly what each label has meant historically.

Diabolical. A long-running newspaper and puzzle-book label for the top tier, alongside "fiendish" in British papers. In most collections a diabolical sudoku maps to what this article calls extreme: at least one advanced pattern needed, usually several.

Evil. Popularized as the fourth and hardest tier on early web sudoku sites, and still widely used by apps that copied that scale. In practice it sits close to expert or low extreme, and is often easier than a puzzle rated "extreme" elsewhere.

Impossible. Almost always a marketing word rather than a claim. A genuine sudoku with a unique solution is by definition solvable by logic, so "impossible" grids are either very deep chain puzzles or, occasionally, badly constructed ones.

That last case is worth taking seriously. If a puzzle is genuinely unsolvable, it is broken, and there are only two ways for that to happen: it has no solution at all, usually from a typo in the givens, or it has more than one, which means at some point no logical deduction exists and the remaining cells can only be filled by choosing. A quick test on a stuck grid is to look for a rectangle of four cells in two rows, two columns, and two boxes holding the same two candidates. That pattern guarantees multiple solutions and means the puzzle, not the solver, is at fault.

Frequently Asked Questions

What is the hardest sudoku puzzle ever made? There's no official title, but the puzzle known as "Everest," constructed by Arto Inkala, is the most widely cited contender, along with forum-famous grids like "Platinum Blonde" and "Golden Nugget."

Is a 17-clue sudoku always the hardest? No. 17 is the proven minimum number of clues for a uniquely solvable puzzle, but many 17-clue grids are only moderately hard, and plenty of puzzles with more clues are considerably more difficult.

Do the hardest sudokus require guessing? No. A properly constructed puzzle always has a complete logical solution path. Extremely hard grids require long chains of conditional reasoning, which feels like guessing but is deduction — each step follows necessarily from the last.

Why do difficulty labels vary so much between puzzle sources? Because each publisher uses its own rating system — some based on the hardest technique required, others on a weighted score across every deduction. One source's "expert" can sit well below another's "hard," which is a property of the rating method rather than of the solver.

How many clues does a typical hard sudoku have? Usually somewhere in the low-to-mid twenties, though placement matters far more than count. Two puzzles with identical clue counts can differ enormously in difficulty depending on where those clues sit.

Is there such a thing as an impossible sudoku? Not among properly constructed puzzles. A grid with a unique solution can always be solved by logic, though the chains required may be far beyond what a human will find in practice. A genuinely unsolvable puzzle is a broken one — either no solution exists, or more than one does.

What does diabolical mean as a difficulty rating? It is a publisher's label for the hardest tier they offer, used mainly in newspapers and puzzle books, and it generally corresponds to what other sources call extreme — puzzles needing X-Wing, swordfish, or chain logic to finish.