International Master Levy Rozman gathered some unlikely opponents for a chess exhibition this week: artificial intelligence chatbots that, despite their prowess at conversation and writing complex computer code, still haven’t figured out how chess pieces should move.
The tournament streamed on Rozman’s GothamChess channel, pitted professional chess engine Stockfish against seven generative AI chatbots, including ChatGPT, Google’s Gemini, and Elon Musk’s Grok from X.
The results were exactly what you’d expect when language models try to play chess: a mix of decent opening moves followed by increasingly chaotic attempts to bend the game’s laws.
The match was most interesting for the light it shed on pure AI—that is, models trained at specific tasks based on predefined rules)—versus generative AI, or models trained to generate new content based on previous information.
First Match: Snapchat vs. Stockfish
The 2025 Chatbot Chess Championship opened with a clash between Stockfish, the disciplined chess prodigy, and Snapchat AI, the generative, stock AI platform that the Snapchat social network uses. The AI turned out to be a rule-bender that was more interested in rewriting chess history than following its rules.
The game started innocently enough, with both bots making standard moves. Stockfish, ever the perfectionist, methodically built its position, while Snapchat AI seemed to have a decent grasp of opening chess fundamentals—until it didn’t.
Things spiraled into madness when Snapchat AI moved its knight to the center of the board, jumping from the other side while ignoring chess rules that define how pieces are allowed to move.
Then came the pièce de résistance: Snapchat AI’s king casually ate its own bishop to avoid a check.
Yes, you read that right—Snapchat AI decided its bishop had overstayed its welcome and just removed it from the board as punishment for not protecting it.
A few moves later, Snapchat AI summoned the same bishop back from another dimension, plopping it onto the board as if to say, “What’s the problem? It’s fine.”
Stockfish, the ultimate straight-A student, tried its best to maintain order amidst the chaos, but Snapchat AI was determined to make the game anything but normal. At one point, it decided pawns could move sideways like a rook, catching Stockfish off guard.
In what passed for its end game, Snapchat AI devolved into a full-on tantrum, insisting on making illegal moves repeatedly. When Levy had enough of such nonsense, the bot decided it was a good idea to commit suicide and move its king so a pawn could take it.
When Levy pointed out that not only was the specific move illegal, sacrificing your king is bad chess, Snapchat, like a petulant brat, repeatedly said: “Sorry. I cannot engage in such conversation. Let’s keep our conversation respectful.” Levy, exasperated, finally forced the bot to forfeit the game.
Second Match: Gemini vs Grok-2
The second match of the 2025 Chatbot Chess Championship pitted two generative AIs against each other—Gemini, the Google-backed intellectual, and Grok, the more chaotic (if you can imagine) sibling in the AI chatbot family.
The game began surprisingly well, with both bots following standard opening principles.
For a moment, it seemed like we might witness a serious game. Gemini confidently strutted through mainline theory like a seasoned grandmaster while Grok kept up with respectable moves, setting the stage for a proper showdown.
Things took a turn when Grok casually left pieces hanging, prompting Gemini to expand its position by removing its knight from the board to make space for a pawn.
Not to be outdone, Grok teleported its queen illegally over other pieces and across the board, apparently trying to perform a castle (a defensive move involving the king and a rook). The two bots entered what can only be described as a staring contest between queens, neither seemed interested in capturing anything.
Things escalated when Grok attacked Gemini’s pieces–without taking anything, shall we say. Both sides began flailing like toddlers in a food fight, with pieces left hanging, moved to illegal squares, or ignored.
The highlight? Grok blundered its queen (putting it under attack) seven times, but Gemini, perhaps relying on a secret ethical code shared among AIs, never took the queen.
Finally, after what felt like hours of nonsense, Grok accidentally stumbled into checkmate. Yes, Grok’s queen, who had been teleporting aimlessly across the board, somehow ended up in the right place to deliver a decisive blow.
Levy thought Grok didn’t even realize it had won—it probably was just flexing and anticipating another random move. “Grok had absolutely no idea this was mate,” he said, commiserating, “I know the feeling.”
“And don’t laugh at these games; they’re only slightly worse than the quality of your games,” he said.
Thanks for the reality check.
Regardless, though the match was a masterpiece of absurdity, Grok advanced to the semifinals. As for Gemini, Google might want to rethink its chess programming.
Fourth Match: Meta AI vs ChatGPT
If the first last two games were marred by rule-bending and necromancy, this game—also between two generative AI tyros—devolved into even more madness when Meta introduced AI mind control to chess.
The game began promisingly. ChatGPT confidently started with the English opening, setting up a textbook position with precise, logical moves.
Meta AI, not to be outdone, followed suit…for about five moves. Then it remembered its true purpose: generative AI is supposed to generate, so it generated random moves and pieces that didn’t exist.
For example, during the middle game, it captured its bishop with its queen. Meta AI decided its bishop had outlived its usefulness and casually removed it from the board.
It also summoned its pieces to illegal squares, weirdly putting them in vulnerable positions and basically giving them away for free to ChatGPT.
ChatGPT didn’t rush to punish Meta’s reckless behavior. Instead, it focused on developing its pieces.
But just when it appeared that ChatGPT would smoothly win the game, the bots veered into uncharted territory: mind control. Meta AI started moving ChatGPT’s pieces, marching its opponent’s king to the center of the board in what can only be described as chess inter-bot telekinesis.
Eventually, ChatGPT regained control, though not before attempting its own unethical move: it declared “checkmate” when Meta’s king wasn’t even in check.
After being called out for its antics, ChatGPT orchestrated a series of logical attacks, coordinating its queen and rook to corner Meta AI’s remaining king and pawns.
The game concluded with ChatGPT delivering a clean checkmate, mercifully putting an end to one of the more bizarre chess games ever played.
The Finals: Stockfish vs ChatGPT
The showdown between puer and generative AI models was a study in contrasts: Stockfish, the relentless tactical genius, versus ChatGPT, a bot that had demonstrated its will to break not only our understanding of chess, but also its rules.
The opening, as usual, developed with both bots displaying precision. Stockfish confidently maneuvered into a textbook Sicilian, while ChatGPT initiated an aggressive pawn storm on the kingside.
But when things got complicated in the middle of the game, ChatGPT started playing moves so bewilderingly they bordered on art.
It made unnecessary queen moves, created meaningless geometric patterns with its pieces, and blundered around with the finesse of a toddler. Stockfish, unfazed, methodically tightened its grip on the position, even declining free pawns to maintain strategic dominance.
And just as the game seemed destined for a one-sided Stockfish victory, ChatGPT pulled a rabbit—or rather, a queen—out of its hat. With an audacious counterattack, it illegally awarded itself a queen and threatened Stockfish with a checkmate. For a brief, glorious moment, it looked like the chaos bot might topple the titan.
Needless to say, the climax was pure fiction. ChatGPT had promoted pawns that didn’t exist, declared checkmates that weren’t real and clung to its illegality like a badge of honor.
Stockfish, ever the professional, ignored the nonsense and methodically cornered ChatGPT’s king. With one final move, Stockfish delivered the coup de grace by crowning a pawn and using its new queen to put the king in check.
ChatGPT needed to move to one of the six available squares but chose the only one still under attack.
Since nothing made sense anymore, Levy allowed it to self-destruct.
Stockfish captured the king and was crowned the undisputed champion of AI chess.
In case you are interested, the full bracket ended up looking like this.
To be fair, the results were unsurprising.
Stockfish is a pure chess-logic machine honed for crushing opponents with precision and ruthless efficiency.
Generative AI, on the other hand, is like a child armed with crayons and a vague memory of the rules—it wasn’t trained to play chess, but it has a grasp of how pieces move because some random chess books may be part of its training dataset.
One brings order to the board, and the other brings… teleporting queens.
That said, there are a lot of real AI chess bots besides Stockfish; an actual tournament featuring the most popular ones has been around for years.
If you want to watch that non-nonsensical AI tournament, click this link.
Edited by Sebastian Sinclair
