ChessBench: A New Benchmark Tests Whether LLMs Can Generate Chess Engines, Not Just Play Chess

A new benchmark asks models to write functional chess engines from scratch — a creative twist on evaluating code generation and algorithmic reasoning.