Play the engine

This is an AlphaZero-style network trained from scratch on extinction chess — no checkmate, and you win by capturing every piece of any one type. A GPU is allocated to your game for as long as you’re playing, and the engine searches while you think. You can pick which training checkpoint to face: az_iter1300 is the one running by default, and the three earlier ones were each the deployed engine in their turn.

New to the variant? The rules differ in one way that changes everything.