← Back to News arXiv · March 2026

Grounded Chess Reasoning: a small model that plays correctly and explains clearly

Engines are like master craftsmen who cannot teach: accurate, but unable to explain. Master Distillation gives a 4B model concise puzzle commentary that surpasses its teacher. The research page carries the full story: background, method, key figures, and links to the paper.

Research pagearXivPromo copy
4BQwen3 base
48.1%puzzle accuracy
+7.2 ppRLVR over SFT
~178tokens per solution
~100×more compact than GPT-5

Chess engines know the best move but cannot explain it; large models are articulate but often play badly. Master Distillation combines both strengths in one 4B model.

Launch Highlights

  • Problem：Chess engines know the answer but cannot explain; large models can explain but often do not know chess.
  • Method：Stockfish supplies truth, Gemini verbalizes the trace, and a student learns verifiable reasoning.
  • Finding：C1-4B reaches 48.1% accuracy and beats the Gemini-3-Flash verbalizer.
  • Why it matters：Wherever a reliable expert program exists, the same recipe can inject expertise into compact models.

Continue reading

The research page covers the background, method, key figures, and paper links; for quick sharing, use the illustrated promo copy.

Open research pageOpen promo copy