Turkish Chess Channel Tests Claude Fable on a Mate-in-Two and Gets the Right Answer

The Turkish chess channel Sabo Satranç opens with a pointed line: "Generative AI systems like ChatGPT, Gemini, and Claude have always failed at chess puzzles, until today." Among AI observers, the tendency of leading LLMs to collapse on tactical chess positions has long been treated as a known weakness.

This time the model under test is Anthropic's new Fable. The host gives it a mate-in-two puzzle from the white side and tests how far it can get toward a proper solution when paired with Stockfish. According to the video, Fable arrives at the correct answer.

The highlight of the segment is not merely that Fable found the move, but that it could explain in natural language why the tactic works. The presenter reconstructs the position's logic by quoting Fable's explanation.

The explanation goes roughly like this: the key is to create a situation in which Black's pieces are functionally unable to move. The knight is pinned in place by the queen's diagonal, while the king, rook, and bishop are all frozen by the surrounding structure.

By locking down the rook and the remaining pawn as well, Black is driven into a zugzwang in which a move is required but every move loses. The only piece left with any legal movement is the a-pawn, and whatever Black does with it, White finishes the mating net with Ra6. That is the logic Fable presents.

As a news item, the timing is especially striking: the video was published immediately after the US government's order to suspend Fable, which gives it value as a record of how the model performed just before it was taken offline. It also captures, from the perspective of Turkish-speaking chess fans, how far LLMs have advanced in expressing tactical chess reasoning in language.