Maia Chess is a research project from the University of Toronto and Microsoft Research that produces a family of chess engines trained to play like human players at specific rating levels (from 1100 to 1900 Elo). Rather than playing the objectively best moves, Maia predicts what a human of a given strength would most likely play.

This makes Maia uniquely valuable as a training opponent: playing against Maia-1500, for example, means facing believable 1500-level mistakes, oversights, and plans rather than the alien, technically perfect play of Stockfish. The opening trainer on the Maia website also uses this approach to show what human opponents are likely to play in response to your repertoire.

Maia is free and open source (MIT license). The engine weights can be downloaded and run locally in any UCI-compatible GUI. An online interface at maiachess.com lets you play directly in the browser and explore the opening trainer without any installation.