What would a 1250 play?

This uses Maia-3, the latest rating-aware model from the University of Toronto's Computational Social Science Lab. It was trained on real Lichess games to predict the exact move a human will play, given both players' ratings. It's the most accurate public model for this.

No model can be 100% accurate, though. Two different 1250 players, or the same player on a different day, will often choose different moves in the same position. Rating-aware models like Maia pick the exact move the real player made roughly half the time. That's close to the ceiling, because human play is partly random. That's why the full probability list is shown. "Play a realistic move" samples from it, the way a real 1250 player would vary.

Ratings refer to the Lichess rating pool. Everything runs locally in your browser.