Stockfish

Stockfish started in 2008 as a fork of Tord Romstad's Glaurung engine, adapted by Romstad, Marco Costalba, and Joona Kiiski. Costalba is usually credited as the founding developer, and the first release date is recorded as 2 November 2008. There is no commercial publisher. The engine is distributed free under the GPLv3 license, with source code hosted on GitHub, and development is driven by a volunteer community of several hundred contributors coordinated by maintainers.

The engine uses alpha-beta search with pruning and reduction heuristics to explore the game tree, combined with an evaluation function that scores leaf positions. For most of its history that evaluation was hand-crafted, with rules for material, pawn structure, piece mobility, and king safety. In August 2020, version 12 introduced NNUE, a hybrid neural network that mostly used the network but could fall back on hand-crafted rules. By July 2023 the hand-crafted evaluation was removed entirely, leaving a fully neural-network-based engine. On strong hardware it searches well over 500 million positions per second.

Stockfish 15 arrived in April 2022 as a free, cross-platform engine used by grandmasters and major chess platforms. Stockfish 17 delivered roughly a 46-point Elo gain over its predecessor and is available for free online analysis through services like Chessify. The most recent major release, Stockfish 18, is documented on the project's official blog in 2026, and community videos peg its strength around 4060 Elo. On the CCRL rating list, Stockfish holds an estimated Elo of about 3650 at 40/15 time controls as of October 2026, the strongest CPU engine rated there. It has won every main event of the Top Chess Engine Championship and the Chess.com Computer Chess Championship since 2020.

There are no Metacritic-style review scores because Stockfish is not a commercial product. Its reception is measured in Elo and tournament titles. Within the chess world it is treated as the benchmark for computer chess, the reference point against which other engines and human play are judged. Its open-source model has had a broad cultural effect: it democratized top-level analysis, made engine cheating a persistent concern in online and over-the-board chess, and even entered general tech discourse. In June 2026 Elon Musk predicted AI would reach Stockfish-level coding by year's end, using the engine as shorthand for superhuman performance. A widely shared 2026 demonstration showed a $299 Ray-Ban Meta smart-glasses setup being used for chess cheating, a controversy that exists precisely because engines like Stockfish are free and ubiquitous. [edit]

Stockfish has no story, graphics, characters, or player-facing presentation. It is a free, open-source computer chess engine: a program that takes a board position through the Universal Chess Interface and returns a calculated best move. The gameplay is the search itself, millions of positions evaluated per second to score candidate moves. It uses alpha-beta search with a long list of pruning and reduction heuristics (null-move pruning, late move reductions, futility pruning) to explore the game tree efficiently, combined with a static evaluation function that scores leaf positions. For most of its history that evaluation was hand-crafted, with rules for material, pawn structure, piece mobility, and king safety written by programmers. In August 2020, version 12 introduced NNUE (efficiently updatable neural network), a hybrid that used a neural net for most evaluation but could fall back on the hand-crafted rules. By July 2023 the hand-crafted evaluation was removed entirely, leaving a fully neural-network-based engine. Stockfish runs on desktop (Windows, macOS, Linux) and mobile (Android, iOS), and on low-end hardware it still comfortably outplays any human; on strong hardware it searches well over 500 million positions per second. It has no built-in multiplayer; multiplayer exists only in the graphical interfaces and platforms that host it, where players can pit Stockfish against humans or against other engines. [edit]