wozitdev's 3D Chess · public · first seen
wozitdev's 3D Chess went into Early Access on 10 November 2025, and the work hasn't stopped since. Nearly eleven months of it lands with this update. Here's what the year adds up to, from the board's side.
The game plays the FIDE Laws of Chess the way an arbiter would enforce them. Touch a piece and you move it (say "J'adoube" first if you only want to straighten it). Offer a draw with your move; claim one on a threefold repetition or the fifty-move rule – on the position as it stands, or with the move that makes it, declared and not played. When a flag falls, it counts only once someone claims it, and the clock can run as FIDE's classical games do: 90 minutes for 40 moves, then 30 more, with 30 seconds a move. Resign, and your king topples – real physics, not an animation. In an online game with people watching, one of the spectators can be made arbiter, with the power to rule on a dispute and hand out the penalties the rulebook allows.
New to chess? The Guide, on the main menu, is a chess handbook inside the game. Learn to Play starts from nothing: the board and how its squares are named, each piece met on its starting square, then check and checkmate, stalemate, castling, en passant, promotion, how a game ends and how to read moves. The Laws, at the back, explain how the game applies the FIDE rules: touch-move, offering and claiming a draw, resigning, the clock and the arbiter. Short pages, one idea to a page, in all 32 languages.
404 puzzles from the Lichess puzzle database, every one a moment from a real game. A puzzle opens by playing its game up to that moment on the board; then it's yours to find the winning line, under the same rules as any game. The list runs from easiest to hardest and teaches as you climb: it starts with mates in one, and each idea – forks, pins, skewers, back-rank mates – comes back later, harder. A wrong move turns red and is taken back. Hint shows the piece to move, then where it goes; Reveal plays the right move for you. Solved puzzles are ticked, per Steam account, and Windowless Board takes them to the desktop too.
The Puzzle Builder, on the Workshop menu, makes your own. Pick the moment from one of your games or any PGN – never a board set up by hand – and play the winning line. The CPU checks that the first move is clearly the best, and the builder estimates the puzzle's rating as you go. Share it on the Workshop; the puzzles you subscribe to appear on the Puzzles screen beside the built-in ones.
Every player now has a wallet of Dollar Dollars ($$), kept on your Steam account so it follows you to any computer. Click an opponent's portrait in the lobby to put a bounty on their head: a sniper reticle hangs over them on every screen until somebody beats them and collects. Lose with a bounty on you and the invoice comes due, and until it's paid the game keeps you quiet – no chat, no GG, no handshake. The arbiter's fines land on the same Invoices list. Sixty is the most a player can carry: after that, a bounty click puts a pair of sunglasses on you instead, worn until you win.
Three builders ship with the game, and everything they make goes to the Steam Workshop:
Browse, rate and subscribe without leaving the game. Play with any set – online, yours shows on your side of the board and your opponent's on theirs. Make a set with no king at all and the game is won by taking every piece. The computer plays every variant by its rules.
The CPU opponent comes in levels, and three of them learned from people. Fish, Easy and Normal were trained on real games from the Lichess database, each on players of its own strength, so they make a person's mistakes, not an engine's. Hard searches ahead and is the strongest of them; Random picks a level for you. Give the CPU a clock of its own if you want it under pressure, and seat it on either side – or both, and watch.
Then there's the Ghost: a CPU made from you. Play ten games and your Ghost is made: your openings, your habits, your pace, even when you'd resign. Options shows how often it plays the move you would, and can pause or reset it. Online it wears your portrait, so friends can play your Ghost, or watch two Ghosts play each other.
Host a public or private game over Steam, invite a friend from the overlay, or join by code. The lobby has two seats and a bench: whoever joins takes a free seat or waits on the bench, and the host can sit out and run a game between two guests or two CPUs. Start Game counts down from three, and can be called off. A game seats up to 100 spectators, who see the handshake, the board and the chat, and a dropped player has a minute to rejoin. Every game goes out as PGN to any chess site or program, and comes back in the same way. And a security review has gone through online play and the Workshop: the game now checks what other players and downloaded items send before it acts on it.
Standard, Rapid, Blitz, Chess960 and Armageddon are built in, each with its own set of pieces. Pick a theme and your board colours; pieces land with a knock, captured pieces fly to the side of the board, and the launch track gives way to an ambient one. A finished game plays back move by move. Moves can be written with the pieces' figures – ♞f3 rather than Nf3 – the default in every language, since the letters change from one language to the next. Not one for handshakes? Turn on Handwave and you wave instead. And Windowless Board: pull the board out of its window and it sits on your desktop, see-through and click-through around the pieces, on top of everything if you like – a game in the corner while you work.
The whole game plays on a pad – every menu, the board, the builders – through Steam Input, with a layout for every controller type, Steam Deck included, and PlayStation glyphs on a PlayStation pad. And the game speaks 32 languages: the 31 Steam offers, and Filipino.
Linux and macOS have had their first real test pass, on the current mainline version of each. The Linux build was installed and played on a native Ubuntu machine, and with no Mac on hand, a cloud-hosted Mac made sure the macOS build launches and plays. Going in, there was no knowing it would all work – only confidence in Godot's exports – and Godot didn't disappoint: every check passed on both.
Windows 11 stays the main platform, so Linux and macOS may trail a step behind it. The game itself looks set to run the same everywhere; what's left to find is anything particular to one system. To match what was tested, the Early Access requirements now name macOS Tahoe and Ubuntu. If you play on either and something's off, say so in the discussions – it helps a lot.
A campaign was on the list from early on, and the work toward it kept turning into other parts of the game. CPU vs CPU, where two computers play it out while you watch, came from it; so did the puzzles and the Puzzle Builder. Each split off and became a system of its own, and the campaign itself stayed unfinished.
This is where it ended up. Campaign now sits at the top of the main menu, with two ways in. Competitor is greyed: there may still be room for a more classic campaign one day, and that's where it would go. Researcher is the campaign as Early Access leaves it: the Chess Solver.
It starts where you can do it yourself. Tic-tac-toe (noughts and crosses) first: prove by hand that O never loses, choosing O's answer to every move X can make, then solve all 5,478 of its positions by computer. Then checkers (draughts): solve the 4×4 board outright and the endings of 6×6, a board proved a draw in 2017. Then chess. Each stage has its goals, from a first game to a solution.
Nobody knows how a perfectly played game of chess ends. What is known is how far solving it has come, one board at a time: 3×3 and 3×4 solved completely, every 4×4 position with up to ten pieces, Gardner's 5×5 minichess proved a draw from its start in 2013, and on the real board every position with seven pieces or fewer, with the first eight-piece tables released in February. On 6×6, Los Alamos chess, the first chess-like game a computer played, in 1956, is unsolved: one line of it was proved in a paper published in 2024, a win for Black in 21 moves when White, playing without a queen, opens 1.d3. The Chess Solver brings you up to that point, keeps up as the work moves on, and gives you the tools to take part on 6×6, with 5×5 still open beside it:
These are the real methods, at the scale a game can run. The frontier takes far more: that Los Alamos proof needed several CPU centuries. But every proof here is checked before it's said, or says it isn't, and everything saves in formats other chess programs read, so you can see for yourself how chess gets solved.
Boards of other sizes have no place in a game built on standard chess and the FIDE Laws, and they still don't: they're here only as the road to solving 8×8, never as variants to play for their own sake. Solving chess is as far as chess goes, so that's where the campaign went.
The one-year anniversary is next month, on 10 November. Thank you for playing through the first year – see you at the board.