Komodo Dragon 3 Released by KomodoChess

[lkaufman]

Komodo Dragon 3 is now available at komodochess.com. It is a significant improvement over the last release 2.6.1, about 17 elo based on 30,000 direct games (20 elo in MCTS mode, 6000 games) on one thread with normal openings at 2' + 1", about double that with unbalanced openings (UHO22 book by S. Pohl). The improvement is due to new net structure, new nets, and search improvements. Compared to the last major version, Dragon 2 (May 2021), the single thread gains (measured by 18000 direct 2' + 1" games with normal openings) are about 107 elo, with the MCTS gain at 132 elo (3,000 games) and the FRC(Chess960) gain at 148 elo (9,000 games). Single thread elo gains by same criteria from the last pre-Dragon Komodo (14.1) came out at 306 elo for standard chess, 377 elo for FRC (4,000 to 5,000 game samples).

Elo levels have been much more carefully tuned and tested, especially it has been verified that Elo 2600 is a good match for Grandmaster Alex Lenderman (who had that FIDE rating) in 15' + 10" Rapid games. The play has been made more human-like at these Elo settings; at the higher levels there should now be few really obvious blunders. UCI Elo functioning has been restored to the method used in 2.6 rather than 2.6.1 by removing the underscore from the name, as the underscore caused problems.

Pricing for Dragon 3 is the same as it was for 2.6.1 (both for single copy and subscription), with older versions going on sale and with Komodo 13.02 now
free. As always recent purchasers get a 20% discount.

[Andrew]

Great news, thanks Larry and Mark!

Andrew

[pohl4711]

Komodo Dragon 3 is now available at komodochess.com.

Why is there still no Android-binary (Armv8, which we can use in Droidfish for example) in the download? For a commercial release, this is not good. Stockfish 15 has it. And is free.

[Guenther]

...

Pricing for Dragon 3 is the same as it was for 2.6.1 (both for single copy and subscription), with older versions going on sale and with Komodo 13.02 now
free. As always recent purchasers get a 20% discount.

Just a little note: The download link for now free Komodo 13 has a little typo in it.

Under
https://komodochess.com/downloads.htm
it points to
http://komodochess.com/pub/komodo-1.zip
instead of
http://komodochess.com/pub/komodo-13.zip

[Dariusz]

Komodo Dragon 3 is now available at komodochess.com. It is a significant improvement over the last release 2.6.1, about 17 elo based on 30,000 direct games (20 elo in MCTS mode, 6000 games) on one thread with normal openings at 2' + 1", about double that with unbalanced openings (UHO22 book by S. Pohl). The improvement is due to new net structure, new nets, and search improvements. Compared to the last major version, Dragon 2 (May 2021), the single thread gains (measured by 18000 direct 2' + 1" games with normal openings) are about 107 elo, with the MCTS gain at 132 elo (3,000 games) and the FRC(Chess960) gain at 148 elo (9,000 games). Single thread elo gains by same criteria from the last pre-Dragon Komodo (14.1) came out at 306 elo for standard chess, 377 elo for FRC (4,000 to 5,000 game samples).

Elo levels have been much more carefully tuned and tested, especially it has been verified that Elo 2600 is a good match for Grandmaster Alex Lenderman (who had that FIDE rating) in 15' + 10" Rapid games. The play has been made more human-like at these Elo settings; at the higher levels there should now be few really obvious blunders. UCI Elo functioning has been restored to the method used in 2.6 rather than 2.6.1 by removing the underscore from the name, as the underscore caused problems.

Pricing for Dragon 3 is the same as it was for 2.6.1 (both for single copy and subscription), with older versions going on sale and with Komodo 13.02 now
free. As always recent purchasers get a 20% discount.

Fantastic News !
Glad to see Dragon being developed so well

[lkaufman]

...

Pricing for Dragon 3 is the same as it was for 2.6.1 (both for single copy and subscription), with older versions going on sale and with Komodo 13.02 now
free. As always recent purchasers get a 20% discount.

Just a little note: The download link for now free Komodo 13 has a little typo in it.

Under
https://komodochess.com/downloads.htm
it points to
http://komodochess.com/pub/komodo-1.zip
instead of
http://komodochess.com/pub/komodo-13.zip

Fixed now. We have a new webmaster, who is doing a great job!

[BrendanJNorman]

Okay, well I've gotta say I'm impressed.

This is only a fast time control game, but the *style* of win is very nice and is a great positional effort by Dragon - no matter how much the tactical wizard Stockfish squirmed, Dragon kept a vice grip on the position.

Things to watch for:

1. Lovely positional pawn sac c5!!
2. Bxe5! and doubling rooks on d-file, followed by knight maneuver to c4 is very human-looking and resembles planning.

[pgn][Event "1 Minutes/Game + 1 Seconds/Move"]
[Site "Elite AF, DESKTOP-C9L0ID5"]
[Date "2022.04.28"]
[Round "1.5"]
[White "Dragon 3 by Komodo Chess 64-bi"]
[Black "Stockfish 15"]
[Result "1-0"]
[ECO "A20"]
[PlyCount "119"]
[EventDate "2022.??.??"]

1. c4 {book 0s} e5 {book 0s} 2. g3 {book 0s} Nf6 {book 0s} 3. Bg2 {book 0s} Nc6
{book 0s} 4. Nf3 {book 0s} Bc5 {book 0s} 5. O-O {book 0s} d6 {book 0s} 6. e3 {
book 0s} O-O {book 0s} 7. Nc3 {0.64/20 3} a6 {book 0s (Bb6)} 8. b3 {0.80/22 7s
} Re8 {0.50/24 10} 9. d3 {0.76/22 3} Bf5 {0.37/24 5s} 10. h3 {0.79/22 2} Ba7 {
0.37/25 5 (Ne7)} 11. Bb2 {0.81/20 4} Rb8 {0.38/21 1} 12. Kh2 {0.75/20 2} b5 {
0.37/21 1 (Nb4)} 13. Nd5 {0.87/23 4} bxc4 {0.39/22 1s} 14. dxc4 {0.93/21 1} Bd7
{0.55/25 6 (Nb4)} 15. Rc1 {0.93/21 3 (Rb1)} h6 {0.62/24 5 (a5)} 16. a3 {
1.07/21 1s} Nh7 {0.55/19 1 (Nxd5)} 17. b4 {1.15/20 2 (Nd2)} Ne7 {0.53/20 1} 18.
Nc3 {1.09/20 1} Bc6 {0.52/21 1 (Ng6)} 19. Qc2 {1.03/22 5} Qc8 {0.45/21 1 (Nf6)}
20. e4 {0.88/20 4} Ng6 {0.42/21 3 (Bd7)} 21. Rfd1 {0.90/21 2s} Bd7 {0.49/18 1}
22. Qe2 {0.76/18 1} Nf6 {0.40/24 2s (a5)} 23. Rd2 {0.83/20 2} Qb7 {0.41/23 4
(Qd8)} 24. Nd5 {0.83/22 3} Nxd5 {0.41/22 1} 25. exd5 {0.76/22 1s} a5 {0.52/22 2
} 26. b5 {0.79/21 1} Bf5 {0.52/22 3 (f5)} 27. a4 {0.97/22 2 (Rdd1)} Bb6 {
1.13/27 12 (Bc5)} 28. Ba3 {1.33/22 2} Qa7 {1.06/19 1 (Rbd8)} 29. c5 {1.35/22 1}
dxc5 {1.04/21 1} 30. Bb2 {1.63/24 3} e4 {1.19/20 1} 31. Ne1 {1.36/23 1} Rbd8 {
1.06/23 2} 32. Nc2 {1.46/22 1} Ne5 {1.23/20 1} 33. Bxe5 {1.24/24 6s (Ne3)} Rxe5
{0.78/19 1} 34. Ne3 {1.33/23 3} Bg6 {0.86/22 2} 35. Rcd1 {1.27/23 2} Bh5 {
1.34/25 6 (Rd6)} 36. g4 {1.72/22 1} Bg6 {1.40/18 0} 37. Nc4 {1.62/21 1} Ree8 {
1.14/22 1 (Re7)} 38. Qe3 {1.87/19 0s} Rd7 {1.33/20 1 (Qb7)} 39. f4 {1.88/21 2
(Qf4)} f6 {1.70/16 0 (f5)} 40. h4 {2.20/18 1} Qb8 {1.98/20 2 (Bf7)} 41. Bh3 {
2.60/18 0} Qd8 {1.89/15 0 (Rde7)} 42. g5 {2.69/19 1} fxg5 {2.27/19 1} 43. hxg5
{2.53/21 2s} hxg5 {2.37/18 0} 44. Be6+ {2.62/23 2 (fxg5)} Rxe6 {1.60/18 0} 45.
dxe6 {2.30/21 0} Rd4 {1.37/19 0} 46. Rxd4 {3.29/25 2 (fxg5)} cxd4 {2.13/20 1}
47. Nxb6 {3.60/25 1} gxf4 {3.26/18 1} 48. Qxf4 {3.48/21 1} cxb6 {3.21/18 0} 49.
Rg1 {3.87/21 1} Be8 {2.88/21 1} 50. Qxe4 {3.45/23 1} Qf6 {3.14/24 1} 51. Rg3 {
3.98/22 1s} Qf2+ {3.14/19 1} 52. Rg2 {3.69/22 0} Qf6 {3.49/17 0s} 53. Kg1 {
4.04/24 2} d3 {3.44/18 0} 54. Qd5 {3.95/21 1} g6 {3.49/18 0 (Qe7)} 55. Rf2 {
4.05/19 0} Qa1+ {3.71/16 0} 56. Rf1 {5.69/22 1 (Kg2)} Qb2 {5.09/18 2 (Qxa4)}
57. Qd8 {8.74/25 1 (Qd6)} Kg7 {4.50/17 0} 58. Qe7+ {9.25/24 0} Kh6 {6.75/22 1}
59. Qf6 {9.49/25 0} Qe2 {7.00/24 1 (Qxf6)} 60. Rf4 {10.36/24 0 (Rf3)} 1-0

[/pgn]

Edit: @Larry, I forgot to mention, was a tiny bit disappointed that the (much-spoken of, by users) variable delay (or what MByrne calls "sleep" in Honey) the lower Elo levels wasn't added yet. It's unnerving for the engine to respond instantaneously to every move, and kills the "real" experience a bit. Impressed the the overall weakening though, there's def improvement there.

[Kirk]

Hi!

What would be the recommended parameters if I wanted to create a "MATEFINDER" personality?

Thanks!

[lkaufman]

Hi!

What would be the recommended parameters if I wanted to create a "MATEFINDER" personality?

Thanks!

That would require some research/testing. I don't even know whether NNUE is a benefit or not for this purpose, you could try "Use Regular Eval" and compare for mating problems. Probably getting Reduction right would be critical for this purpose. If you do "Use Regular Eval", then parameters like King Safety and Dynamism also become significant. They don't work with NNUE on.

[Kirk]

Hi!

What would be the recommended parameters if I wanted to create a "MATEFINDER" personality?

Thanks!

That would require some research/testing. I don't even know whether NNUE is a benefit or not for this purpose, you could try "Use Regular Eval" and compare for mating problems. Probably getting Reduction right would be critical for this purpose. If you do "Use Regular Eval", then parameters like King Safety and Dynamism also become significant. They don't work with NNUE on.

Thank you!