In December 2017, DeepMind’s AlphaZero stunned the chess world. After just a few hours of self-play training, it defeated the reigning champion engine Stockfish. Its sacrificial, dynamic style looked more like a daring human grandmaster than a traditional computer. Soon after, the open-source project Leela Chess Zero brought this revolution to the public, ushering in a new neural-network era of chess engines.
AlphaZero often sacrificed material for dynamic compensation: long-term initiative, dark-square control, or unstoppable pawn storms. Grandmasters described it as “alien chess”, with ideas rarely seen in engine play.
Classical engines like Stockfish calculate tens of millions of moves per second. By contrast, AlphaZero searched only ~80,000 positions per second, relying on its neural network to guide moves. This was a complete paradigm shift:
AlphaZero demonstrated a near-human sense of strategy. Games often featured knight outposts, pawn sacrifices for initiative, and long-term king hunts — plans that traditional engines previously undervalued.
Because AlphaZero’s code was never released, the chess community responded by creating Leela Chess Zero (Lc0), an open-source engine inspired by AlphaZero’s principles.
Leela reached the TCEC Superfinals multiple times, even defeating Stockfish in individual games with stunning sacrifices reminiscent of AlphaZero’s style.
Some critics argued the conditions of the Stockfish match were unfair, with suboptimal settings. Others noted that newer Stockfish versions quickly closed the gap. Nevertheless, AlphaZero had changed the conversation forever.
The AlphaZero & Leela era redefined computer chess. It proved that creativity and intuition can be programmed, and that AI can teach humans new ways to think about the game. Today, every serious chess player — amateur or professional — studies the games of these engines to expand their understanding.
👉 Continue exploring the evolution of chess in our Chess History Guide.