Best Chess Openings for Engines: What 335,111 Engine Games Show (2023-2026)

The short answer

• With no opening book, the top engines played the Catalan and the Ruy Lopez (most often the Berlin) in 784 of 1,208 games (64.9%). 95.8% of those games were draws.
• In real tournaments the engines almost never choose. 99.1% of all 335,111 games started with book moves picked by the organisers.
• The Sicilian Defence is the most played family by far: 71,270 games, 21.3% of the total.
• In CCRL games, the King's Indian gave the most decisive games (35.2%) and the Grünfeld the fewest (23.8%).

So the "best opening for engines" depends on what you want. For safety, the top engines pick the Catalan and the Berlin. For a decisive game, organisers pick sharp, unbalanced positions. The details and the limits are below.

First, a warning: engines rarely choose their openings

Two top engines that both play the best move every time would play the same game again and again, and almost every game would be a draw. So engine tournaments give them an opening book: a list of start positions, usually 6 to 12 moves deep. Each position is normally played twice, once with each engine as White. The book moves are marked "book" in the game files, so we could count them:

• CCC (Chess.com Computer Chess Championship): book moves in 100.0% of games; the book is almost always exactly 8 moves (16 half-moves). White scores 72.8%.
• TCEC (Top Chess Engine Championship): book moves in 94.1% of games; a typical book is 6 moves (12 half-moves), some much longer. White scores 64.6%.
• CCRL broadcast games: book moves in 99.2% of games; usually 8 or 12 moves. White scores 55.2%.

Look at the White scores. In a normal start position White scores only a little more than half. CCC and TCEC choose books where one side, usually White, has a clear plus, so that games are not all draws. That is the most important thing to know before reading any table on this page: the openings and their results say as much about the organisers' books as about the engines.

What top engines play with no opening book

A few TCEC bonus events let the engines start from the normal position with no book. We found 1,208 such games between top engines (2023-11-07 to 2026-04-24). This is the closest thing we have to the engines' own choice of opening.

TCEC bonus events from 2023 to 2026 that started from the normal position with no opening book (no book move in the game file). Pairs: Stockfish vs Lc0 (804 games), Komodo Dragon vs Berserk (200), Stockfish vs Reckless (100), Stockfish vs Integral (100), plus 4 self-play games. Most of these events were bullet or blitz. Families with 20+ games shown.

What stands out:

• 1.e4 (595 games) and 1.d4 (598) were played almost equally often. Other first moves: only 15 games.
• With White, Stockfish mostly played 1.e4 and got the Ruy Lopez (275 games) or the Italian (102). Lc0 mostly played 1.d4 and got the Catalan (342 games).
• With Black, Stockfish met 1.d4 with ...e6 and ...d5, which gave Catalan positions (382 games). Lc0 met 1.e4 with 1...e5 and reached the Ruy Lopez (241 games).
• 95.8% of the games were draws. White scored 50.1%, almost exactly half.

Limits: most of these events were bullet or blitz, two thirds of the games are Stockfish vs Lc0, and engines repeat themselves: the 1,208 games contain only 211 different first-8-move sequences. So read it as "what two or three engines liked in 2024-2026", not as a law.

Watch: Stockfish 19 out of a quiet opening

On our channel Stockfish 19 plays a new engine. A quiet opening, then a rook sacrifice. Watch it, then see the numbers below.

The most played openings in engine games

Now all 335,111 games. The Sicilian is far ahead, mainly because the CCRL book uses it so often (31.0% of CCRL games). In CCC the Sicilian has 13.0%, in TCEC 12.1%.

Share = games in that family / all games of that source. Opening named by position with the Lichess opening list (the deepest named position in the first 30 half-moves). In CCC, TCEC and CCRL almost every game starts from a book chosen by the organisers, so this table shows what the organisers picked, not what the engines prefer.

By first move: 1.e4 started 156,702 games and 1.d4 115,443. 1.Nf3 and 1.c4 follow with 26,873 and 19,517. For the human side of these openings, see our chess openings list and the opening tier list.

Which openings give the most decisive engine games?

To compare openings fairly we need games where the book is not built to push one side. The CCRL games are the best fit: White scores 55.2% there, not far from a normal start. Here are the families with the most and the fewest decisive games.

CCRL broadcast games, 2025-11-02 to 2026-10-30, families with 1,500+ games (19 families). Decisive = White wins + Black wins. CCRL plays many engines of very different strength, so the level is the same for every opening, but it is not the level of the top 5 engines.

The pattern is clear. Openings with unbalanced pawn structures, where both sides attack (King's Indian Defense, Dutch Defense, French Defense), give more decisive games. Solid queen's-pawn openings (Slav Indian, Queen's Indian Defense, Grünfeld Defense) give the most draws. The gap is real but not huge: 35.2% vs 23.8%.

Draw rate by year, 2023 to 2026

Engines get stronger every year, and stronger engines draw more. But the CCC draw rate went down, from 57.5% in 2023 to 49.3% in 2026. At the same time White's score went up from 69.8% to 74.8%. The likely reason is the book: CCC moved to openings that favour White more, to keep games decisive. In TCEC long games the draw rate stayed between 51% and 54%.

White score = White wins + half the draws. TCEC: only games with a long time control (about 30 minutes or more per side), so every year is the same kind of game. CCRL games in our archive start on 2 November 2025.

A book opening that decides the game

This is what an organiser's book looks like in practice: the TCEC Superfinal between the two strongest engines. Both games of this opening pair were won by White.

How we measured

• Games: 335,111 games from 821 PGN files: CCC event files from Chess.com (149,982 games), the official TCEC archive (29,874 games) and CCRL broadcast games (155,255 games, from 2025-11-02). Played from 1 January 2023 to 30 October 2026, read on 12 November 2026.
• Left out: 9,188 Fischer Random, Double Fischer Random and odds games; 578 games outside the dates; 13 games with no result or no moves.
• Opening names: by position, with the Lichess opening list. We take the deepest named position in the first 30 half-moves, so move orders do not matter. The family is the name before the colon ("Sicilian Defense: Najdorf" = Sicilian Defense). 2,695 games could not be named (most start from a set position).
• Book moves: counted from the "book" comments in the files. The no-book table uses TCEC bonus events with no book move at all, between top engines (tiny 4K engines and handicapped test builds left out).
• Time controls are mixed: CCC is mostly bullet and blitz, TCEC mostly long games; the CCRL files do not give the time control. That is why we split by source where it matters.
• Board evaluations: Stockfish 18, depth 16, 2 threads. Board moves are copied from the official game files.
• Scripts: engine_openings_scan.py, engine_openings_compute.py, engine_openings_boards.py, build_engine_openings.py. Every number on this page is in engine_openings.json.

Our other data page: chess engine sacrifice statistics (which engine sacrifices the most). More about the events: TCEC explained.

Engine Opening Statistics FAQ