What are the best engines against humans in chess?

[Uri Blass]

Even if all humans lose it is possible to decide that the engine who get winning position in smaller average number of moves is the best.

We cannot be sure if a position is winning but we can dedice that if the best engine can beat itself then the position is probably winning.

I also wonder what is the average number of moves that engines need to get a winning position with white or with black against different level of rating of humans.

If top humans played in the usual ambitious way they use to play against each other, I don't think there would be a very noteable difference between various engines performance-wise or if there were one it would be mostly coincidence.

The real question is how this would look if humans tried to play a bit like "Father", without any other ambition than to draw games and/or drag them on.

I think top human players would even get a pretty decent number of draws against Stockfish this way, the same would be much more difficult against Chess System Tal EAS.

My interest is because I would like a computer to help me to prepare opening book with practical chances to win as fast as possible against humans (at level of fide rating near 2000) and I think the moves that an engine that wins faster choose are probably better for that purpose relative to stockfish's moves.

I wonder what is the smallest size of book that you need to get a winning position out of book against humans at different levels in most games with white.

[jefk]

PB wrote

I don't think there would be a very notable difference between various engines performance-wise

well i doubt that; just played a few test/simulation games, see below, the aggressive
engine (Reb-eas21) as i expected is quickly winning (with my gambit book) against an
engine simulating human/club level play. Probably faster than SF would do.


[pgn][Event "Computer Chess Game"]
[Site "DESKTOP-G2238OJ"]
[Date "2025.05.20"]
[Round "1"]
[White "Chess System Tal 2.1 EAS"]
[Black "Chenard 2023.06.23"]
[Result "1-0"]
[TimeControl "300+2"]

1. Nf3 Nf6 2. e4 Nxe4 3. d4 d5 4. Bd3 Nc6 5. O-O Bg4 6. c4 f5 7. Qa4 Bxf3
8. gxf3 Nf6 9. Bxf5 dxc4 10. Nc3 a6 11. Be6 Qxd4 12. Rd1 Qb6 13. Nd5 Nxd5
14. Rxd5 g6 15. Bd7+ Kf7 16. Qxc4 Kg7 17. Be3 Qxb2 18. Rc1 Qf6 19. Bxc6
bxc6 20. Bd4 cxd5 21. Bxf6+ exf6 22. Qxc7+ Kh6 23. Qf4+ Kh5 24. Qg4+ Kh6
25. Qh4+ Kg7 26. Rc7+ Kg8
{Xboard adjudication} 1-0
[/pgn]

[pgn][Event "Computer Chess Game"]
[Site "DESKTOP-G2238OJ"]
[Date "2025.05.20"]
[Round "2"]
[White "Chenard 2023.06.23"]
[Black "Chess System Tal 2.1 EAS"]
[Result "0-1"]
[TimeControl "300+2"]

1. d4 Nf6 2. c4 c5 3. dxc5 e6 4. b4 a5 5. Qa4 Na6 6. a3 axb4 7. axb4 Nc7 8.
Qxa8 Nxa8 9. Rxa8 Be7 10. g3 b6 11. cxb6 Bxb4+ 12. Bd2 Qxb6 13. Rxc8+ Ke7
14. Rxh8 Ba5 15. Bxa5 Qxb1+ 16. Kd2 Qa2+ 17. Kd3 Qxa5 18. e3 Qf5+ 19. Kd2
Ne4+ 20. Ke2 Qxf2+ 21. Kd3 Nc5+ 22. Kd4 d6 23. Kc3 Qxf1 24. h4 Qc1+ 25. Kb4
{Xboard adjudication} 0-1
[/pgn]

[pgn][Event "Computer Chess Game"]
[Site "DESKTOP-G2238OJ"]
[Date "2025.05.20"]
[Round "3"]
[White "Chess System Tal 2.1 EAS"]
[Black "Chenard 2023.06.23"]
[Result "1-0"]
[TimeControl "300+2"]

1. e4 e5 2. d4 exd4 3. f4 Nc6 4. Nf3 d5 5. e5 Bb4+ 6. Nbd2 Bg4 7. Bd3 f6 8.
h3 Bd7 9. a3 Be7 10. Qe2 fxe5 11. fxe5 Bh4+ 12. Kd1 Bg3 13. Nb3 Qc8 14. e6
Bxe6 15. Nbxd4 Nd8 16. Ng5 Ne7 17. Ngxe6 Nxe6 18. Nxe6 g6 19. Bg5 Bd6 20.
Re1 Rg8 21. Rc1 Qd7 22. Qf3 a6 23. c4 c6 24. c5 Bb8 25. Qf6 Be5 26. Rxe5 a5
{Xboard adjudication} 1-0

[/pgn]Why i choose chenard, instead of eg ufim which also can simulate human play
https://www.perplexity.ai/search/what-i ... IRiNXgpQ#1

Now to confirm the idea that aggressive engines faster against moderate humans
than SF (as apparently also thought by UriW and CW and possibly ES aka rebel), I
should also test this Chenard engine against Stockfish but i leave that to the SF boys


PS when talking about the 'father' style, well this is precisely why i choose my
gambit book to play with RebelEAS, in such a way applying an ant-comp style
as 'father' usually is doing, is most likely much more difficult (cq impossible).

PS2 fyi (for those not yet knowing) found this latest RebelEas21 at the outskirt forum
where i already happened to have an account (otherwise you cant download)
https://tinyurl.com/5hfy6x7p

[jefk]

should also test this Chenard engine against Stockfish

Now started a few test games Chenard vs SF17 , adjudicated at move 31 with score >5

[pgn][Date "2025.05.20"]
[Round "1.1"]
[White "Stockfish 17"]
[Black "Chenard 2023.06.23"]
[Result "1-0"]
[TimeControl "600"]

{ERT 6.0.4.4 Autoflag=On Pondering=Off OperatorTime=300 mS } 1. d4 Nf6 2. c4 e6
3. Nf3 b6 4. g3 Ba6 5. Qc2 Nc6 6. Bg2 Nb4 7. Qb3 d5 8. a3 Bxc4 9. Qa4+ b5
10. Qd1 Nc6 11. b3 Be7 12. O-O O-O 13. Bb2 Rb8 14. Nbd2 g6 15. bxc4 bxc4
16. Bc3 Rb6 17. Re1 Ra6 18. e4 Qe8 19. exd5 exd5 20. Ne5 Nd8 21. Qe2 Ne6
22. Ndxc4 dxc4 23. d5 Nd8 24. Qxc4 Rb6 25. a4 Nd7 26. Qxc7 Rb7 27. Ng4 h5
28. Qc4 Rb6 29. Nh6+ Kh7 30. a5 Rd6 31. Bb4 Nb7 {Score threshold reached} 1-0
[/pgn]

Now setting adjucation limit at 1600 pawn value, on averaged C will last at least 30
moves i predict, which is less than what i observed on average against RebelEas21
(at least with my gambit book).

Also played a quick test game myself against SF17 and I also could last 30 moves.

Both P4 and RebelEas are more difficult to play against at ordinary human level
because they will merciless start an attack, with usually a kings attack later and
you need enormous defensive skills to avoid being mated before move 30.

In the end it doesn't matter much, i admit, whether you get into a completely lost
endgame or being checkmated, but for me it's clear that higher (engine)-Elo doesnt
mean faster wins against (most) humans. Elo of P4 or Rebel-eas21 are good enough
and give more interesting chess (games); certain imo with a gambit book

[Uri Blass]

PB wrote

I don't think there would be a very notable difference between various engines performance-wise

well i doubt that; just played a few test/simulation games, see below, the aggressive
engine (Reb-eas21) as i expected is quickly winning (with my gambit book) against an
engine simulating human/club level play. Probably faster than SF would do.


[pgn][Event "Computer Chess Game"]
[Site "DESKTOP-G2238OJ"]
[Date "2025.05.20"]
[Round "1"]
[White "Chess System Tal 2.1 EAS"]
[Black "Chenard 2023.06.23"]
[Result "1-0"]
[TimeControl "300+2"]

1. Nf3 Nf6 2. e4 Nxe4 3. d4 d5 4. Bd3 Nc6 5. O-O Bg4 6. c4 f5 7. Qa4 Bxf3
8. gxf3 Nf6 9. Bxf5 dxc4 10. Nc3 a6 11. Be6 Qxd4 12. Rd1 Qb6 13. Nd5 Nxd5
14. Rxd5 g6 15. Bd7+ Kf7 16. Qxc4 Kg7 17. Be3 Qxb2 18. Rc1 Qf6 19. Bxc6
bxc6 20. Bd4 cxd5 21. Bxf6+ exf6 22. Qxc7+ Kh6 23. Qf4+ Kh5 24. Qg4+ Kh6
25. Qh4+ Kg7 26. Rc7+ Kg8
{Xboard adjudication} 1-0
[/pgn]

[pgn][Event "Computer Chess Game"]
[Site "DESKTOP-G2238OJ"]
[Date "2025.05.20"]
[Round "2"]
[White "Chenard 2023.06.23"]
[Black "Chess System Tal 2.1 EAS"]
[Result "0-1"]
[TimeControl "300+2"]

1. d4 Nf6 2. c4 c5 3. dxc5 e6 4. b4 a5 5. Qa4 Na6 6. a3 axb4 7. axb4 Nc7 8.
Qxa8 Nxa8 9. Rxa8 Be7 10. g3 b6 11. cxb6 Bxb4+ 12. Bd2 Qxb6 13. Rxc8+ Ke7
14. Rxh8 Ba5 15. Bxa5 Qxb1+ 16. Kd2 Qa2+ 17. Kd3 Qxa5 18. e3 Qf5+ 19. Kd2
Ne4+ 20. Ke2 Qxf2+ 21. Kd3 Nc5+ 22. Kd4 d6 23. Kc3 Qxf1 24. h4 Qc1+ 25. Kb4
{Xboard adjudication} 0-1
[/pgn]

[pgn][Event "Computer Chess Game"]
[Site "DESKTOP-G2238OJ"]
[Date "2025.05.20"]
[Round "3"]
[White "Chess System Tal 2.1 EAS"]
[Black "Chenard 2023.06.23"]
[Result "1-0"]
[TimeControl "300+2"]

1. e4 e5 2. d4 exd4 3. f4 Nc6 4. Nf3 d5 5. e5 Bb4+ 6. Nbd2 Bg4 7. Bd3 f6 8.
h3 Bd7 9. a3 Be7 10. Qe2 fxe5 11. fxe5 Bh4+ 12. Kd1 Bg3 13. Nb3 Qc8 14. e6
Bxe6 15. Nbxd4 Nd8 16. Ng5 Ne7 17. Ngxe6 Nxe6 18. Nxe6 g6 19. Bg5 Bd6 20.
Re1 Rg8 21. Rc1 Qd7 22. Qf3 a6 23. c4 c6 24. c5 Bb8 25. Qf6 Be5 26. Rxe5 a5
{Xboard adjudication} 1-0

[/pgn]Why i choose chenard, instead of eg ufim which also can simulate human play
https://www.perplexity.ai/search/what-i ... IRiNXgpQ#1

Now to confirm the idea that aggressive engines faster against moderate humans
than SF (as apparently also thought by UriW and CW and possibly ES aka rebel), I
should also test this Chenard engine against Stockfish but i leave that to the SF boys


PS when talking about the 'father' style, well this is precisely why i choose my
gambit book to play with RebelEAS, in such a way applying an ant-comp style
as 'father' usually is doing, is most likely much more difficult (cq impossible).

PS2 fyi (for those not yet knowing) found this latest RebelEas21 at the outskirt forum
where i already happened to have an account (otherwise you cant download)
https://tinyurl.com/5hfy6x7p

You use both gambit book and different engine.
It may be interesting to compare also with only one change relative to stockfish without gambit book.

Also it may be interesting to compare number of moves that you need to get a winning position and not number of moves to mate.

[Uri Blass]

I do not think that your engine is good in simulating humans at level of 2000

I see
1. d4 Nf6 2. c4 c5 3. dc5 e6 4. b4

I looked at statistics of database of human games from chesstempo.

Almost nobody play 4.b4 even at level of 2000 or lower and only stupid materialistic engines may play it.

[jefk]

do not think that your engine is good in simulating humans at level of 2000

In the example i didn't want to simulate humans at level of 2000, but below at, average
club level, thus around 1600, and it were only two games. Other example, RebelEas plus
gambit book against humanoid engine Ufim, also (like chenard) at approx 1600:

[pgn][Event "Computer Chess Game"]
[Site "DESKTOP-G2238OJ"]
[Date "2025.05.20"]
[Round "5"]
[White "Chess System Tal 2.1 EAS"]
[Black "Ufim 8.02 weak=20"]
[Result "1-0"]
[TimeControl "300+2"]

1. e4 e5 2. Nc3 Nf6 3. f4 exf4 4. e5 Ng8 5. d4 Qh4+ 6. Ke2 Ne7 7. Nf3 Qg4
8. h3 Qg3 9. Qd3 d5 10. exd6 cxd6 11. Nb5 Nd5 12. Qe4+ Be6 13. c4 g6 14.
cxd5 Na6 15. dxe6 fxe6 16. Qxe6+ Kd8
{Xboard adjudication} 1-0
[/pgn]

At Elo level 2000, i'm not going to repeat such experiments (with other engines, see below); P.Berger may
be right that SF on average is winning as fast as R-Eas or P4; with sharper openings, another parameter, also
faster wins may be obtained.
NB at such levels of 2000 Elo, almost Fide master, you (or other strong human players) in trying to see
how you would do against SF or P4 could watch engines as Myrrdin or Rodin, set at such
levels in the ini file. They are included in the latest Chess Gambiteer package at :

https://sourceforge.net/projects/chess-gambiteer/files/

[flora17]

Thanks for these advices, I'm looking for an engine too!