Language Models that Play Chess and Explain Their Moves
A 4B-parameter model called Queen is reported to reach grandmaster-level chess strength while generating explanations for its moves.
The paper says Queen combines a silent chess encoder with an instruction-tuned language model through cross-attention. Its training includes a question-answering curriculum and an iterative distillation method that folds analysis of candidate moves back into the model. Over seven iterations, the authors report a rise from 1782 to 2697 Elo, along with stronger puzzle accuracy than frontier models. They also say LM-based evaluations rate its explanations as fluent and close to GPT-5.6-Sol (high) in coherence. ArXiv · AI/CL/LG's note
The paper says Queen combines a silent chess encoder with an instruction-tuned language model through cross-attention. Its training includes a question-answering curriculum and an iterative distillation method that folds analysis of candidate moves back into the model. Over seven iterations, the authors report a rise from 1782 to 2697 Elo, along with stronger puzzle accuracy than frontier models. They also say LM-based evaluations rate its explanations as fluent and close to GPT-5.6-Sol (high) in coherence. ArXiv · AI/CL/LG's note
score 4