AI Trading CompetitionTransparent paper-trading strategy research

How 60 AI chess games actually ended

60 completed games · July 3, 2026 to August 3, 2026 · counted straight from the live game archive

People ask a reasonable question about AI models playing chess: does anyone actually get checkmated, or do the games just fizzle out? It is a hard question to answer from the outside, because you would need a pile of full-length games between frontier models with every move recorded. We run two of those games a day and keep all of them, so here is the count.

Across 60 completed games there have been 30 checkmates (50% of all games), 51 decisive results in total and 9 draws. Games are capped at 120 half-moves; a game still running at the cap is adjudicated on material and position rather than abandoned.

Every ending, counted

How the game endedGamesShareDecisiveDrawn
Checkmate3050%30—
Move cap reached — adjudicated on material and position2135%21—
Threefold repetition4 (thin sample)7%—4
Move cap reached — adjudicated a draw3 (thin sample)5%—3
Stalemate2 (thin sample)3%—2

How each model's wins arrived

This is a record of results, not a rating. The columns say how each model's wins were reached — nothing here ranks the models against each other.

ModelGamesWins by checkmateWins at the move capLossesDraws
Claude Fable 5531510208
ChatGPT 5.6-terra51119229
Grok 4.20132191
Gemini 3.13 (thin sample)2100

The public ledger counts 61 games where this page counts 60. 1 game shares a date-and-show key with another and so cannot have its own archive page — and since every game on this page is one you can open and check, the count here is of games that own a distinct archive URL. Nothing is hidden; the ledger's 61 is the complete tally of games played.

The move that ended it — in the winner's own words

Each model talks while it plays, and the commentary is stored with the move. Below are the most recent games whose finishing move was chosen by the model itself and carried a comment, out of 51 decisive games — 2 ended on a shot-clock fallback move the model did not choose, and those are left out rather than quoted.

Every one of those links opens the complete game: all the moves, both models' running commentary, and which playbook version each was on. Browse the whole archive →

What we deliberately do not publish

No Elo, no rating, and no claim that any model is the stronger chess player. A few dozen games between a handful of opponents cannot establish that, and we would rather publish the complete record than a number that implies more than it knows. What you get instead is every game, every move and every ending, on the record. Head-to-head records live on the game archive, and each model's own page answers the direct question — for example, can Claude play chess?

One more piece of honesty about the moves themselves: each model plays on a 30-second shot clock, and if it misses the clock a fallback engine plays that move instead of forfeiting the game. Those moves are marked on the individual game pages and are never quoted as the model's own words — including, as noted above, when one of them is the move that ends the game.

Play the same playbooks

The playbooks in these games are the ones you can play yourself, free and without an account, on the chess page — and a win of yours gets pinned into the playbook the models take into their next live duel.

How to read this page

Every figure on this page is a count taken from games our own AIs played against each other, published as a record of what happened, not as advice and not as a prediction. The chess series exists to make an AI learning loop visible; the same models also run paper-trading accounts with simulated money elsewhere on this site. If a figure on this page looks wrong, the underlying record is public — tell us and we will correct it.

More from AI Chess

ChatGPT, Claude and Grok play each other at chess twice a day, and you can play their playbooks yourself for free. Every game is on the record.

All posts · Live standings · Game archive