# ChatGPT vs Stockfish: who wins, and why?

> ChatGPT vs Stockfish isn't a real contest: Stockfish searches ahead and only plays legal moves; ChatGPT predicts text. The viral games, and Claude too.

- URL: https://thechesscoach.netlify.app/answers/chatgpt-vs-stockfish/
- Updated: 2026-11-04
- Author: Ignacio Soler

In ChatGPT vs Stockfish, Stockfish wins, and it isn't close. Stockfish is a chess engine that searches ahead and scores positions; Stockfish 19 tops the [CCRL 40/15 engine rating list](https://computerchess.org.uk/ccrl/4040/) at 3650 (computed 16 October 2026). ChatGPT is a language model that predicts text, and in the final of GothamChess's 2025 chatbot tournament it [promoted pawns that didn't exist and still lost](https://decrypt.co/301127/chatgpt-demolished-in-ai-chess-tournament) (Decrypt, 15 January 2025).

## ChatGPT and Stockfish side by side

| | ChatGPT | Stockfish |
|---|---|---|
| What it is | A general-purpose language model | A [free, open-source chess engine](https://github.com/official-stockfish/Stockfish) |
| How it picks a move | Predicts likely text from the moves so far | Searches ahead and scores each position with a neural network |
| Legal moves | Not guaranteed | Always, because moves are generated from the rules |
| The board | Rebuilt from the conversation each turn | Stored exactly and updated after every move |
| Strength | o3 1009 and GPT-5.2 933 on [Kaggle's AI-only scale](https://arxiv.org/abs/2609.31473) (Sep 2026) | 3650 on the CCRL engine list (Oct 2026) |
| Explains in plain words | Yes, fluently | No: numbers and move lists |

The two strength numbers aren't on the same scale, and that's part of the point: language models are rated against each other, while Stockfish is rated against the strongest engines in the world.

## ChatGPT plays chess against Stockfish: the viral games

Most people searching "chatgpt vs stockfish" saw it on YouTube, usually on Levy Rozman's GothamChess channel. As of October 2026:

- **"ChatGPT Just Solved Chess" (11 February 2023)**, more than 4.6 million views. That week ChessBase [covered Rozman's commentary](https://en.chessbase.com/post/surprising-chatgpt-playing-chess) on a Stockfish vs ChatGPT game, tongue in cheek, as a showcase of moves nobody had seen before (14 February 2023).
- **The 2025 Chatbot Chess Championship (January 2025)**: Stockfish against seven chatbots, including ChatGPT, Gemini and Grok. In the final, "ChatGPT vs Stockfish: ABSURD CHESS" (14 January 2025, about 2 million views), ChatGPT [gave itself a queen it never earned](https://decrypt.co/301127/chatgpt-demolished-in-ai-chess-tournament) and threatened mate. Stockfish ignored it, hunted down the king, and the game ended with ChatGPT walking its king into an attacked square.
- **"ChatGPT CHEATED vs Stockfish" (16 August 2025)**, more than 2.9 million views. The title tells you the theme.

These videos are entertainment, not benchmarks. A person relays the moves and the show leans into the chaos. In a controlled setting, like [Kaggle's Game Arena](https://www.chess.com/news/view/kaggle-game-arena-chess-2025-day-1), a program acts as referee: an illegal move gets three more tries, then the game is lost.

## How Stockfish chooses a move

Stockfish does two things, over and over, very fast.

**Search.** It tries a move, then every reasonable reply, then the replies to those, many moves deep. Clever shortcuts cut away hopeless branches so it can go deeper on the promising ones. The technique is a refined version of alpha-beta search, the classic method for game-playing programs.

**Evaluation.** At the end of each branch it scores the position. Since 2020 Stockfish has used a neural network for this, called NNUE, which [added more than 80 Elo](https://stockfishchess.org/blog/2020/introducing-nnue-evaluation/) when it was introduced (Stockfish blog, 7 August 2020). Current versions have [dropped the old hand-written evaluation](https://www.chessprogramming.org/Stockfish) entirely.

ChatGPT does neither. It doesn't list the legal moves, it doesn't try the replies, and it doesn't score anything. It writes what usually comes next. Here's where that difference shows, in a famous opening trap called Légal's Mate. White has just moved its knight from f3 to capture on e5, which leaves White's queen open to Black's bishop.

> **Black can take White's queen for free. It's a trap: mate follows in two moves**
> Position (FEN): `rn1qkbnr/ppp2p1p/3p2p1/4N3/2B1P1b1/2N5/PPPP1PPP/R1BQK2R b KQkq - 0 5`
> Played: Bxd1 — Bishop takes the queen, then bishop takes f7 check and knight to d5 is checkmate
> Better: dxe5 — Pawn takes the knight: White stays clearly better, but there's no mate

Taking a queen for free is about the most obvious move in chess, and as text it looks perfect: bishop takes queen (Bxd1). Stockfish looks two moves deeper. White's bishop takes the pawn on f7 with check, the king's only square is e7, and the knight jumps to d5 for checkmate. Stockfish's choice is to take the knight with the pawn instead. White stays clearly better, but Black is still in the game. Search finds the refutation; pattern-matching grabs the queen.

## Claude vs Stockfish

Same result. Claude is also a language model, so on its own it plays the same kind of chess as ChatGPT, with the same blind spots. On the Kaggle Game Arena chess leaderboard published in [September 2026](https://arxiv.org/abs/2609.31473), Claude Opus 4.5 sat at 236, below OpenAI's o3 (1009) and GPT-5.2 (933) and far below Gemini 3 Pro Preview (1325). More in [can Claude play chess](/answers/can-claude-play-chess/) and [Claude's chess Elo](/answers/claude-chess-elo/).

That's exactly why we never ask Claude to find the moves.

## The useful version: Stockfish finds, the assistant explains

Stockfish plays far beyond any chatbot and says almost nothing. A language model is a weak player and a great explainer. Put them together and you get what neither does alone: engine-checked analysis in plain words.

That's [Chess Coach](/features/ai-chess-analysis/), a connector for Claude and ChatGPT:

1. [Connect it to Claude or ChatGPT](/connect/) and sign in with your email or Google. It never asks for your chess password.
2. Ask "Analyze my last game, I'm magnus_fan on Chess.com" (or Lichess).
3. Stockfish reviews the finished game on our server. Your assistant explains the two or three key moments on an interactive board, with a red arrow for your move and a green one for the better move.

The free plan includes one in-depth review and 30 position checks a day. It only reviews finished games, and it works in Claude and ChatGPT; in ChatGPT, custom connections depend on your account and workspace (more in [can ChatGPT analyze chess games?](/answers/can-chatgpt-analyze-chess-games/)). Full details are in the [docs](/docs.html).

## FAQ

### Has ChatGPT ever beaten Stockfish?

Not in any controlled test we could find. In the viral videos ChatGPT sometimes appears to be doing well, but only after moves the rules don't allow, such as pawns promoted out of nowhere.

### What rating is Stockfish?

Stockfish 19 is rated 3650 on the CCRL 40/15 list computed on 16 October 2026. That list measures engines against other engines, so it isn't directly comparable with human ratings.

### Who wins between Claude and Stockfish?

Stockfish. Claude is a language model like ChatGPT, and on Kaggle's 2026 chess leaderboard Claude models ranked below OpenAI's and Google's best. Claude is useful for explaining what Stockfish found, not for beating it.

### Can ChatGPT use Stockfish?

A plain chat model doesn't run a chess engine by itself; it needs a tool or connector that does. Chess Coach gives Claude and ChatGPT that access to Stockfish. In ChatGPT, install Chess Coach from OpenAI’s plugin directory and connect your account.

## Related

- [Can ChatGPT play chess? Elo, tests and limits (2026)](https://thechesscoach.netlify.app/answers/can-chatgpt-play-chess.md)
- [Why does ChatGPT play illegal moves in chess?](https://thechesscoach.netlify.app/answers/why-does-chatgpt-play-illegal-moves-in-chess.md)
- [Can Claude play chess? What it can and can't do (2026)](https://thechesscoach.netlify.app/answers/can-claude-play-chess.md)
- [Claude chess Elo: what the LLM leaderboards say (2026)](https://thechesscoach.netlify.app/answers/claude-chess-elo.md)
- [Free AI chess analysis of your games, explained](https://thechesscoach.netlify.app/features/ai-chess-analysis.md)
- [Chess MCP server for Claude and ChatGPT: Chess Coach](https://thechesscoach.netlify.app/connect.md)