# Grok chess: can Grok play, and how well?

> Grok chess, honestly: Grok plays legal moves, but it trails Gemini, OpenAI and Claude on public LLM chess leaderboards and lost the 2025 Kaggle final 4–0.

- URL: https://thechesscoach.netlify.app/answers/can-grok-play-chess/
- Updated: 2026-10-25
- Author: Ignacio Soler

Grok can play chess, but not very well for a 2026 chatbot. xAI's reasoning models know the rules and
almost always play a legal move when they're shown the position. On public LLM chess leaderboards as of
October 25, 2026, though, Grok is well behind Gemini, OpenAI's models and Claude. In chess, Grok is best
known for losing the final of Kaggle's 2025 AI tournament 4–0 to OpenAI's o3.

Here's the Grok chess record on four public tests. Elo is the rating system chess uses: higher means
stronger. Compare numbers only within one column; each board has its own pool and rules.

| Grok model | [AI Chess Leaderboard](https://dubesor.de/chess/chess-leaderboard) (Oct 25, 2026) | [LLM Chess Leaderboard](https://maxim-saplin.github.io/llm_chess/) (data of Oct 20, 2026) | [LLM Chess Benchmark](https://chessbenchllm.onrender.com/) (checked Oct 25, 2026) | [Game Arena paper](https://arxiv.org/abs/2609.31473) (Sept 2026) |
|---|---|---|---|---|
| Grok 4 | 1415 ±118 (retired) | not tested | not tested | 773 |
| Grok 4.20 Reasoning | 1362 ±89 | 843 ±183 | not tested | not tested |
| Grok 4.1 Fast Reasoning | 1096 ±93 (retired) | 800 ±112 | 1442, listed as grok-4.1-fast (retired) | 632 |
| Grok 4.7 | 1063 ±107 | 890 ±217 (only 11 games) | not tested | not tested |
| Leader, for comparison | Gemini 3.1 Pro Preview: 1860 ±85 | GPT-6 Astra (high): 1614 ±110 | Gemini 3.1 Pro Preview (high): 2144 | Gemini 3 Pro Preview: 1325 |

On Dubesor's board, a human-like network trained on games by 1500-rated Lichess players (maia5-1500)
sits at 1554 ±90, above every Grok model, and a bot playing random legal moves sits at 419.

## Grok chess Elo: what the leaderboards show

Grok's chess Elo depends on the board, but the pattern is the same everywhere: Grok sits in the middle
of the pack, below the best models from Google, OpenAI and Anthropic. On Saplin's leaderboard, the best
Grok (Grok 4.7 with high thinking) is at 890 ±217, against 1575 ±113 for Claude Opus 5.5. In the Game
Arena paper, Grok 4 (773) is behind Gemini 3 Pro Preview and o3, but well ahead of Claude Opus 4.5 (236)
from the same period.

Newer isn't always stronger. On Dubesor's board the retired Grok 4 is still Grok's top entry, and
Grok 4.7 (1063 ±107) sits below Grok 4.20 Reasoning (1362 ±89). Thinking matters a lot: Grok 4.20 in
its non-reasoning version is at 589 ±94 on the same board.

Mistakes are the other gap. Dubesor counts a move as a blunder when it cuts the player's winning chances
by 15% or more. Grok 4.3 blundered on 9.7% of its moves, against 2.2% for Gemini 3.1 Pro Preview.
And in Dubesor's continuation mode, where the model only sees the moves played so far, 81% of
Grok 4.20 Reasoning's move attempts are legal, compared with 100% in his reasoning mode.

## The 2025 Kaggle final: o3 4–0 Grok 4

Grok's most famous chess moment came in August 2025, at Kaggle's AI chess exhibition. Grok 4 beat
Gemini 2.5 Flash 4–0 in the quarterfinals, as
[Chess.com reported](https://www.chess.com/news/view/kaggle-game-arena-chess-2025-day-1), then edged
Gemini 2.5 Pro in a deciding game in the semifinal. Before the final, Elon Musk posted that xAI had
spent almost no effort on chess, as [Decrypt reported](https://decrypt.co/334367/sam-altman-openai-elon-musk-grok-ai-chess).

Then o3 won the final 4–0. According to
[Chess.com's report of the final](https://www.chess.com/news/view/kaggle-game-arena-chess-2025-day-3)
(August 7, 2025), Grok dropped pieces in game after game. In game three, Grok had White and a calm,
equal position after ten moves.

> **Grok 4 (White) against o3, game three of the Kaggle final, August 2025**
> Position (FEN): `r1bq1rk1/4ppbp/p1pp1np1/6B1/2P1P3/2N5/PP2BPPP/R2Q1RK1 w - - 0 11`
> Played: Nd5 — Knight to d5 attacks the knight, but a pawn simply takes it
> Better: Qd2 — Queen to d2 develops and keeps the game about equal

In our check with Stockfish, the position before the knight jump is about equal. The knight attacks
Black's knight, but nothing protects it on d5, and a black pawn can take it for free. After that,
Black is completely winning: a whole piece up for nothing. Chess.com's report says Grok went on to lose
its queen and more. A "hanging" piece like this one, a piece that can be taken without paying anything
back, is the same mistake beginners make.

There's a funny footnote. After the semifinal, the coach Peter Heine Nielsen pointed out a mate Grok had
missed, and Grok's account on X replied that it was sharpening its tactics for the final, per the same
[Chess.com semifinal report](https://www.chess.com/news/view/kaggle-game-arena-chess-2025-day-2). It
lost the next four games.

## Is there a Grok chess bot?

We couldn't find an official Grok chess bot or chess mode from xAI as of October 2026. You can still
play Grok in a normal chat:

- **Paste the full move list every turn**, so it doesn't have to remember the board.
- **Keep a real board open**, such as [Lichess's analysis board](https://lichess.org/analysis), and
  check every move it suggests.
- **Use a reasoning model.** On every leaderboard above, Grok's thinking versions play far better than
  the fast ones.

Never use Grok, or any chatbot, during a game you're playing online: Chess.com and Lichess both ban
outside help. For how the other chatbots compare, see
[Claude vs ChatGPT at chess](/answers/claude-vs-chatgpt-chess/) and
[Can Gemini play chess?](/answers/can-gemini-play-chess/)

## Chess Coach and Grok

Chess Coach doesn't work in Grok. It's a connector for Claude and ChatGPT that reviews your finished
Chess.com and Lichess games with Stockfish on our own server. Your assistant then explains the key
moments in plain words on an interactive board, with a red arrow for the move you played and a green
one for a better move, like the knight blunder above. Every evaluation comes from the engine, so the
chatbot's own playing strength doesn't matter. If you use Claude or ChatGPT, the free plan includes one
in-depth review a day: [connect Chess Coach](/connect/), read the [docs](/docs.html), or see
[why chatbots play illegal moves](/answers/why-does-chatgpt-play-illegal-moves-in-chess/).

## FAQ

### Can Grok play chess?

Yes. Grok knows the rules and xAI's reasoning models almost always play a legal move when they're given the position. But it isn't a chess engine, and on October 2026 leaderboards it trails Gemini, OpenAI and Claude.

### What is Grok's chess Elo?

There's no official rating. On Dubesor's AI Chess Leaderboard, Grok 4 is at 1415 ±118 and Grok 4.20 Reasoning at 1362 ±89 (October 25, 2026). These are ratings among AI players, not Chess.com or FIDE ratings.

### Did Grok lose to ChatGPT at chess?

Grok 4 lost the final of Kaggle's AI chess exhibition 4–0 to OpenAI's o3 in August 2025. It had beaten Gemini 2.5 Flash 4–0 and Gemini 2.5 Pro on a tiebreak to get there.

### Is there a Grok chess bot?

We couldn't find an official Grok chess bot or chess mode from xAI as of October 2026. You can play Grok in a normal chat by typing moves, keeping a real board open to check them.

### Does Chess Coach work with Grok?

No. Chess Coach is a connector for Claude and ChatGPT only; it doesn't run inside Grok.

## Related

- [Claude vs ChatGPT chess: which AI plays better? (2026)](https://thechesscoach.netlify.app/answers/claude-vs-chatgpt-chess.md)
- [Can Gemini play chess? Gemini's chess Elo in 2026](https://thechesscoach.netlify.app/answers/can-gemini-play-chess.md)
- [Can ChatGPT play chess? Elo, tests and limits (2026)](https://thechesscoach.netlify.app/answers/can-chatgpt-play-chess.md)
- [Can Claude play chess? What it can and can't do (2026)](https://thechesscoach.netlify.app/answers/can-claude-play-chess.md)
- [Chess MCP server for Claude and ChatGPT: Chess Coach](https://thechesscoach.netlify.app/connect.md)