Can Gemini play chess?

Yes, Gemini can play chess, and it's the strongest chatbot family at it on most public tests. As of October 25, 2026, Gemini 3.1 Pro Preview leads Dubesor's AI Chess Leaderboard at 1860 ±85, ahead of the best OpenAI and Claude models. But Gemini is a language model, not a chess engine: it still makes mistakes, and in 2025 a Gemini model missed a mate in one in a televised AI tournament.

Here's what the public leaderboards show for Gemini. Elo is the rating system chess uses: higher means stronger. Each board uses its own pool and rules, so compare numbers only within a column.

For scale, on Dubesor's board a human-like network trained on games by 1500-rated Lichess players (maia5-1500) sits at 1554 ±90, and a bot playing random legal moves at 419. On Maxim Saplin's LLM Chess Leaderboard (data of October 20, 2026), the best Gemini, Gemini 4 Argon with high thinking, is at 1609 ±131, level with OpenAI's GPT-6 Astra (1614 ±110).

Gemini chess Elo: what the numbers mean

Gemini's chess Elo is an estimate inside each leaderboard, not a human rating. Dubesor rates AI models by the games they play against each other. The LLM Chess Benchmark tries to estimate a Lichess Classical rating using fixed engine and bot anchors. Kaggle's Game Arena paper only ranks models against each other and sets the weakest one, DeepSeek V3.2, to 0.

So how strong is 1860 on Dubesor's board? He gives a rough, low-confidence comparison from his own tests: about 1850 there, the level of the best AI players, matched Chess.com's level-16 bots, which are rated around 2000. That's a decent club player in a slow game. It's nowhere near Stockfish, which the same leaderboard uses as the referee to judge the AI players' moves.

Gemini also blunders less than its rivals. On Dubesor's board, 2.2% of Gemini 3.1 Pro Preview's moves were blunders (moves that cut its winning chances by 15% or more), against 2.8% for Claude Opus 5.5. For the full comparison, see Claude vs ChatGPT at chess.

Can Gemini 3 play chess?

Yes, and it's what put Google at the top. In the Game Arena paper, Gemini 3 Pro Preview (1325) and Gemini 3 Flash Preview (1297) form a top tier that the authors describe as clearly separated from the rest; OpenAI's o3 is next at 1009. The paper also counts "rethinks", the retries a model needs after an illegal move. Gemini models needed fewer than 0.2 per game against most opponents. Weaker models often needed more than five.

The catch is the thinking setting. On the LLM Chess Benchmark, Gemini 3 Pro Preview scores 1953 with high thinking but 623 with medium thinking, and Gemini 3 Flash Preview drops from 1857 (medium) to 650 (low). The fast answer you get by default in a chat app may play much worse than the leaderboard headline.

Gemini still blunders: the mate in one it missed

The best-known example comes from Kaggle's AI chess exhibition in August 2025. Gemini 2.5 Pro and Grok 4 were tied 2–2 in the semifinal and played a deciding game, as Chess.com reported on August 6, 2025. After 13 moves, Gemini (White) had a checkmate in one move.

r . . q k b . r
. . . . n p . p
p . . p . . . .
. p . N p b . .
. . . . . . . .
. . P . . . . .
P P N . . P P P
R . . Q K B . R

In our check with Stockfish, the knight jump to f6 is mate on the spot: it gives check, and Black's king can't escape: its own pieces block it, and the knight covers the last free square. Gemini played the other knight instead. Black traded off the strong knight, and White was still clearly better, but the instant win was gone. Later in the same game, Gemini gave away its queen in a winning position. The game ended in a draw, which was enough for Grok to reach the final.

Gemini knows its limits, at least sometimes. In July 2025, Gemini first claimed it would easily beat the 1979 Atari 2600 Video Chess cartridge, then called off the match once the tester told it about his earlier games against the cartridge, as The Register reported on July 14, 2025.

Can I play chess against Gemini?

Yes. Ask for a game in a normal chat and type your moves. A few things help:

• Share the whole game every turn. Paste the full move list, not just your last move, so Gemini doesn't have to remember the board.
• Keep a real board open. Play the moves on a free board such as Lichess's analysis board to catch illegal moves.
• Use the strongest model with thinking on. The leaderboard numbers come from Gemini's top settings.

In December 2024, Google added a "Chess champ" Gem for Gemini Advanced subscribers, where you play by typing moves like e4, as 9to5Google reported. We couldn't confirm whether it's still offered as of October 2026. And as with any chatbot, never ask Gemini for help in a game you're playing online: Chess.com and Lichess both ban outside help.

Chess Coach and Gemini

To be clear: Chess Coach doesn't work in Gemini. It's a connector for Claude and ChatGPT that reviews your finished Chess.com and Lichess games with Stockfish on our own server, then has your assistant explain the key moments in plain words on an interactive board. Every evaluation comes from the engine, not from the chatbot's guess, so a chatbot's own chess strength doesn't limit the review. If you use Claude or ChatGPT, the free plan includes one in-depth review a day: connect Chess Coach, see how the AI chess coach works or read the docs. For the other chatbots, see Can Claude play chess? and Can ChatGPT play chess?