♟️ Play Chess Against a Reasoning LLM

Interactive demo of the tiny chess "thinking" models from Understanding Reasoning from Pretraining to Post-Training (model: pavelslab-nyu/Chess-SFT-Models · model_680m_48B).

Given the game so far, the model first reasons (chain-of-thought between <T> … </T>), then plays a move. Enter your move in SAN (Nf3, e4, O-O) or UCI (g1f3, e2e4) — or pick from the legal-move dropdown — and the LLM will think and reply. These are small research models (~680M params), so they occasionally reason their way into an illegal move; when that happens the app falls back to a random legal move so the game can continue.