TOP 20:
# PLAYER : RATING ERROR POINTS PLAYED (%) 1 Stockfish 19 : 3803 12 1075.0 1902 56.5% 2 Reckless 0.10.0-dev-7300f044 : 3775 11 1219.0 2280 53.5% 3 PlentyChess 8.0.0 : 3762 11 1347.5 2660 50.7% 4 Integral 8.0 : 3758 13 540.0 1140 47.4% 5 Lc0 0.32.1 BT4-614750 4060m : 3757 20 283.0 480 59.0% 6 Pawnocchio 2.0.1 : 3750 12 1313.0 2660 49.4% 7 Obsidian 16.14 : 3747 10 1473.0 2660 55.4% 8 Viridithas 20.0.0 : 3737 11 1258.5 2662 47.3% 9 Alexandria 9.0 : 3733 10 2529.5 4940 51.2% 10 Stormphrax 8.0.0 : 3729 13 928.0 1900 48.8% 11 Hobbes 3.0 : 3727 13 708.0 1520 46.6% 12 Berserk 14 : 3726 13 707.0 1520 46.5% 13 Raphael 4.2.0 : 3709 14 502.5 1140 44.1% 14 Caissa 1.25 : 3704 12 933.0 1900 49.1% 15 Clover 9.1 : 3703 10 1613.5 3420 47.2% 16 Quanticade 3.0 : 3693 12 763.0 1520 50.2% 17 Halogen 16.0 : 3692 10 1522.0 3420 44.5% 18 Tarnished 5.0 : 3689 11 1107.0 2290 48.3% 19 Horsie 1.1 : 3683 9 2513.5 5702 44.1% 20 Stormphrax 7.0 : 3672 11 1027.0 2280 45.0%
Updates:
Tested Integral 8.0 and Musashi 1.3.
Click here for the complete rating list:
The rating list – updated as of 20/09/2026 – is calculated with Ordo and was obtained under the following assumptions.
- The Elo rating of Rebel 6 UCI (1994) is now revised at 2350 points, 100 points less than before, and serves as an “anchor” for all others. This score is the average value of the results obtained by the original version of the engine in various matches against human GMs in the early 90s, derived from various official or unofficial sources (Talkchess.com, Rebel13.nl, computerchessuk.com, various forums…). I revised the number after new data I’ve found.
- Several electronic chessboards have been added whose strength in terms of Elo obtained through matches with humans is fairly well known. In this way, the rating list should provide engine scores that are more easily comparable with human ones. The electronic chessboards were emulated with CB-Emu, and made to play against several UCI and Winboard engines of similar strength.
- The time per game was set to 40/120′ repeated, reparameterized to the processing speed of a Pentium 90. The execution speed of the latter was emulated, and through various benchmarks done in the past on real P90 machines, I tried to obtain a value as close as possible to reality. Consequently, on modern PCs the actual time per game was 40/125” or 40/130” (seconds) depending on the PC used for the test, comparable to other blitz tests found on the net. For some engines, such as Chess Titans, it was not possible to define a game time; for this reason, I have indicated the characteristics of the CPU on which it was tested.
- The opening suite consists of 190 different positions, repeated for each engine (each engine played the same opening both as white and black). For older engines, not equipped with a UCI or XBoard interface (such as the 1988 version of BattleChess), I manually reproduced the various moves suggested by the programs. Also for these engines, as for all the others, the playing time was reparameterized to the performance of a Pentium 90. For this reason, the games of these engines are few compared to the total, and based on random selections of the 190 openings. The opening suite can be downloaded here.
- All the chess engines used are freely available on the net, with the exception of commercial programs that I purchased years ago, and are limited to using only one CPU core.
- In the “Top 10”, I have not included past versions of Stockfish, nor engines derived from it (such as Shashchess). In the complete rating list, I have also included past versions of Stockfish for comparison.
- The main goal of this rating list, made for my exclusive enjoyment, was both to determine the playing level of modern engines compared to past ones, but also to evaluate the actual playing strength of many programs that I loved in my youth (the aforementioned Battlechess, for example).