No. Chess is not solved, and on any timeline that matters to a living person, it never will be. A game is “solved” when the outcome under perfect play — win, loss, or draw — is known with a proof, and ideally when a perfect move is known for every position that can arise. For chess we have neither. We do not know whether the starting position is a win for White or a draw, we have no proof of either, and the space of positions is so vast that no computer that exists or is plausibly projected could enumerate it. The honest, current answer to “is chess solved” is a flat no, with the qualification that small fragments of the endgame have been solved completely.

The question gets asked because engines now play so far above the strongest humans that it feels like the game must be finished. It is not. Engine strength and a solution are different things, and the gap between them is enormous. This piece gives the direct answer and the numbers; for the deeper game-theory treatment — what “weakly” versus “strongly” solved means, and how a smaller variant was actually cracked — see Antichess Was Solved. Can Chess Be?.

Is chess solved?

No. To “solve” chess in the formal sense used by game theorists would mean knowing the game-theoretic value of the initial position with a proof. The strong consensus among engine analysts is that chess is a draw with best play, but that is a belief grounded in evaluation, not a theorem. Nobody has proved the value of the starting position, and nobody has even solved the first few moves of the game to a terminal result. The frontier of what has been proved sits at the far end of the endgame, not anywhere near the opening.

Has chess been solved before, in any part?

Only the endgame, and only the shallow end of it. Every position with seven or fewer pieces on the board — kings included — has been computed exhaustingly into endgame tablebases. The Syzygy and Lomonosov projects store, for each of these positions, its exact value and a move that achieves it. The six-piece Syzygy set fits in roughly 150 GB; the seven-piece set is about 17 TB. These are genuine solutions: within those regions, the game is finished, and they have already overturned century-old human verdicts on certain endings that turn out to be winnable with sequences of fifty-plus precise moves no human would find over the board.

But seven pieces is a long way from thirty-two. The next tablebase frontier, eight pieces, is estimated at around two petabytes of compressed storage — feasible but expensive. Nine pieces is beyond any single project today. The progression makes the scale of the full problem clear: each piece added multiplies the work by orders of magnitude, and the game starts with thirty-two of them.

What would solving chess require?

It would require touching, or provably ruling out, every position the search cannot dismiss by symmetry or transposition. That is where the numbers become absurd. The number of legal chess positions is estimated at roughly 1044. The number of distinct possible games — the Shannon number, the standard order-of-magnitude estimate published by Claude Shannon in 1950 — is about 10120. For comparison, the number of atoms in the observable universe is around 1080. The game tree of chess is larger than the universe by forty orders of magnitude.

No storage medium and no processor budget that exists or is credibly projected can enumerate 1044 positions, let alone the game tree above it. This is not a problem that faster chips solve. Even the speed-ups offered by quantum search algorithms do not dent the order-of-magnitude barrier; they would shave an exponent in a way that leaves the task no more achievable. The obstacle is the size of the object, not the speed of the tool.

Why hasn’t a strong engine solved it already?

Because strength is not proof. Modern engines — Stockfish with its hand-tuned search and neural evaluation, or Leela Chess Zero with its self-trained networks — choose excellent moves by searching deeply and evaluating heuristically. They do not exhaust the tree. They prune almost all of it. A move that an engine rates as best at depth 60 is a confident estimate, not a proven optimum, and engines still disagree with one another on the assessment of specific lines. New Leela training runs occasionally shift evaluations in positions previously thought settled. That is exactly what an unsolved game looks like: near-optimal play that remains, formally, unverified.

Will chess ever be solved? Can it be?

In principle, yes — chess is a finite game with no hidden information, so a perfect strategy exists and could in theory be computed. In practice, no, not on any timeline relevant to humans. The cleanly solved games are the small ones. Connect Four was solved in 1988. Checkers, the largest game ever solved cleanly, was proved a draw by Jonathan Schaeffer’s team in 2007 after roughly two decades of computation, and checkers has about 1020 positions — twenty-four orders of magnitude smaller than chess. The pattern is consistent: small state-spaces fall to computation, large ones do not, and chess sits far on the wrong side of that line. Go, at roughly 10170 positions, sits further still; AlphaGo and its successors play it superhumanly without coming anywhere near solving it.

So the practical answer to “will chess be solved” and “can chess be solved” is the same: not in any future we can plan around. A genuinely new model of computation, or a mathematical shortcut that nobody has yet glimpsed, would be required — and no such shortcut is on the horizon.

What an unsolved game gives us

The consequence is a strange and rather pleasant one. Chess will go on being played at a level engines call near-optimal while remaining, formally, unsolved. The endgame is partly finished — the Lucena and Philidor positions and their kin are confirmed by tablebase — yet the middlegame is an open frontier explored by neural networks rather than exhausted by brute force. For the player this is liberating. The game is not merely unsolved; it is structurally so much larger than anyone who plays it that the question of a solution never touches the experience of sitting down at the board. Humans will keep winning and losing for as long as the game is played.

If you want the formal vocabulary — weakly solved, strongly solved, ultra-weakly solved — and the story of a variant that genuinely was cracked, the companion piece on antichess goes there in detail.

References

Cross-links inside Caissly: the deeper game-theory treatment is in Antichess Was Solved. Can Chess Be?; related modern variants include Antichess, Atomic, and Chess 960.

Issue Nº 005 · The Magazine · The Caissly Editorial