Rook odds match: human vs engines.

[lkaufman]

After 1300 and something games, Komodo 12 level 17 has +26 Elo against Dragon 2.6 Elo 2000. This is a bit stronger than I predicted. Taking into account that now I am somewhat better than Komodo 12 level 17 at 15+10, and that I am only 1900 at chess.com rapid, there is some confusion. I estimated that Komodo 12 level 17 was about 1800, but it turns out that it's much stronger. There are only two logical answers - either I got qiute stronger than 1900 recently, or Dragon Elo 2000 is overestimated. Somehow I think that the letter is true.

There is a third plausible explanation, which I am inclined to believe. Dragon, using NNUE, is very different from Komodo, using hand-crafted eval. Dragon is dramatically stronger positionally, rarely making strategic/position errors which all HCE engines make. So for Dragon Elo 2000 to play near the level of Komodo 17 level 17 in direct matches, it has to be much weaker tactically (due to a much shorter search depth). Against humans even this shorter search is usually enough, and the vastly better positional play makes all the difference against humans, but in direct matches being outsearched by 3 plies or so is a really big deal. So even though you tend to beat Komodo 12 level 17, you might struggle against Dragon Elo 2000. But I would also say that if you can really make a plus score against Komodo 12 level 17, I think you must be stronger than 1900. I recall that level 17 was a pretty good match for average GMs in slow blitz (5' + 2"), and I can't imagine that a 2500 rated GM playing slow blitz would have much trouble beating a 1900 player playing Rapid. Do you have any other ratings you can quote, such as chess.com blitz, Fide, Lichess blitz or Rapid, etc.? I know they are not directly comparable, but I know how to convert from one scale to another reasonably well.

[Fritz 0]

Larrry, I had made a typo, I meant Komodo 12 level 14, which I edited. I surely can make a solid plus score against it at 15+10 recently. I can even put a solid resistance to it at 5'+5''.

[Chessqueen]

You do NOT need to test it, I am testing Komodo 12.1.1 Skill = 14 and Vs Bell which is rated around 1962. Sayuri 2018.05.23 is rated around 1850, Komodo 12.1.1 was set to Skill = 14

419 Snowy 0.2 64-bit 1996 +26 −26 51.5% −11.9 19.0% 554
53.8%
420 BSC 3.9 1994 +37 −37 45.1% +38.0 24.0% 263
58.8%
421 ALChess 1.84 1989 +23 −23 52.8% −22.3 26.8% 672
68.7%
422 Heracles 0.6.16 1981 +19 −19 47.8% +16.6 26.8% 1034
86.5%
423 Dabbaba 5.67 1963 +25 −25 45.5% +34.0 26.4% 557
53.1%
424 BeLL 0.0.41 64-bit 1962 +22 −22 50.1% −0.8 24.5% 722
64.8%


[pgn][Event "Computer chess game"]
[Site "MININT-UB2PIMJ"]
[Date "2022.04.21"]
[Round "?"]
[White "Komodo-12.1.1-64bit-bmi2"]
[Black "BELL 0_0_41"]
[Result "0-1"]
[BlackElo "1976"]
[ECO "A00"]
[Opening "Dunst (Sleipner-Heinrichsen-Van Geet) Opening"]
[Time "14:25:27"]
[Variation "1...d5"]
[WhiteElo "?"]
[TimeControl "900+10"]
[Termination "normal"]
[PlyCount "111"]
[WhiteType "program"]
[BlackType "program"]

1. Nc3 {(Nb1-c3 e7-e6 d2-d4) +0.16/6 0} d5 {(d7d5) -0.13/11 30} 2. e3
{(e2-e3 d5-d4 e3xd4 Qd8xd4 Qd1-f3 Ng8-f6) +0.48/6 0} e6 {(e7e6) -0.18/10
29} 3. d4 {(d2-d4 c7-c6 Bf1-d3) +0.23/6 0} Bd6 {(f8d6) -0.16/10 28} 4. e4
{(e3-e4 d5xe4 Nc3xe4 c7-c6 c2-c3 Bd6-f8) +0.87/6 0} dxe4 {(d5e4) -0.15/10
28} 5. Nxe4 {(Nc3xe4 Ng8-f6 Ne4xd6+ Qd8xd6 c2-c3 O-O Bc1-d2) +0.46/6 0} Nf6
{(g8f6) -0.18/9 27} 6. Bg5 {(Bc1-g5 c7-c6 c2-c3) +0.77/6 0} Nc6 {(b8c6)
-0.35/9 26} 7. Bxf6 {(Bg5xf6 g7xf6 c2-c3 Bd6-e7 Qd1-d3 Qd8-d5) +0.90/6 0}
gxf6 {(g7f6) -0.13/10 26} 8. c3 {(c2-c3 f6-f5 Ne4xd6+ c7xd6 Qd1-d2 Qd8-a5
Ra1-d1 Qa5xa2) +0.82/6 0} f5 {(f6f5) +0.38/9 25} 9. Nxd6+ {(Ne4xd6+ c7xd6
Qd1-d2 Qd8-h4 Qd2-e2 Qh4-e4 Qe2xe4) +0.47/6 0} Qxd6 {(d8d6) +0.22/9 25} 10.
Qd2 {(Qd1-d2 b7-b6 Ra1-d1) +0.58/6 0} Bd7 {(c8d7) +0.18/9 24} 11. Nf3
{(Ng1-f3 Qd6-f4) +0.37/6 0} Qd5 {(d6d5) +0.10/8 24} 12. Qe2 {(Qd2-e2 h7-h6)
+0.46/6 0} Rg8 {(h8g8) +0.41/9 23} 13. h4 {(h2-h4 h7-h6 Nf3-d2 Rg8xg2
Bf1xg2 Qd5xg2) +0.24/6 0} Rg4 {(g8g4) +0.53/8 23} 14. Ng5 {(Nf3-g5 Nc6xd4
c3xd4 Rg4xd4 Ng5-f3 Rd4-a4) +1.09/6 0} h6 {(h7h6) +0.61/9 22} 15. Nh7
{(Ng5-h7 Ke8-e7 Qe2-d3) -1.06/6 0} Ke7 {(e8e7) +1.47/10 22} 16. Qd2
{(Qe2-d2 Qd5-e4+ Ke1-d1 Rg4xh4 Rh1xh4 Qe4xh4 Qd2-e2) -0.35/6 0} Rh8 {(a8h8)
+1.88/10 22} 17. Qxh6 {(Qd2xh6 Rg4-g6 Qh6-f4 Rh8xh7 f2-f3 Rg6-g7) -2.32/6
0} Rg6 {(g4g6) +1.92/9 21} 18. Qf4 {(Qh6-f4 Rh8xh7 h4-h5 Rg6-g4 Qf4xc7
Ke7-f8 Qc7xb7) -2.76/6 0} e5 {(e6e5) +1.92/10 21} 19. dxe5 {(d4xe5 Rh8xh7
Ra1-d1 Qd5-e4+ Qf4xe4 f5xe4 Rd1-d5) -1.37/6 0} Rxh7 {(h8h7) +2.07/10 20}
20. Rd1 {(Ra1-d1 Qd5xe5+ Qf4xe5+) -2.09/6 0} Qxa2 {(d5a2) +2.53/9 20} 21.
h5 {(h4-h5 Qa2xb2 Qf4-e3) -1.05/6 0} Qxb2 {(a2b2) +2.82/9 20} 22. Qe3
{(Qf4-e3 Rg6-e6 Bf1-c4) -1.81/6 0} Rg4 {(g6g4) +3.08/9 19} 23. f3 {(f2-f3
Rg4-g7 h5-h6) -1.42/6 0} Rg8 {(g4g8) +3.11/9 19} 24. f4 {(f3-f4 Ke7-e8
Qe3-d3) -1.45/6 0} Ke8 {(e7e8) +3.13/9 19} 25. Qd3 {(Qe3-d3 Bd7-e6 Rd1-b1)
-1.76/6 0} Bc8 {(d7c8) +2.84/9 18} 26. Rb1 {(Rd1-b1 Qb2-a2 Rb1-d1 Rg8xg2
Qd3-f3 Rg2-g4) -1.99/6 0} Qa2 {(b2a2) +3.12/9 18} 27. Rd1 {(Rb1-d1 Rg8xg2
Bf1xg2 Qa2xg2 Rh1-f1 Qg2-e4+ Qd3xe4 f5xe4) -2.18/6 0} Qa4 {(a2a4) +3.19/9
18} 28. Rh4 {(Rh1-h4 a7-a6 g2-g3 Bc8-e6 Ke1-f2 Qa4-a3) -1.99/6 0} Be6
{(c8e6) +3.20/8 17} 29. g3 {(g2-g3 Qa4-e4+ Ke1-f2 a7-a5 h5-h6 Rg8-g6
Qd3xe4) -2.04/6 0} f6 {(f7f6) +3.71/9 17} 30. Qe3 {(Qd3-e3 Rh7-d7 Rd1xd7
Ke8xd7 e5xf6 Qa4-e4 Qe3xe4) -2.31/6 0} Qe4 {(a4e4) +3.89/10 17} 31. Qxe4
{(Qe3xe4 f5xe4 Rd1-b1) -3.37/6 0} fxe4 {(f5e4) +4.02/12 17} 32. Kf2
{(Ke1-f2 f6xe5 Rd1-b1 Rh7-g7 f4xe5 Rg8-f8+ Kf2-e3) -2.99/6 0} e3+ {(e4e3)
+4.24/10 16} 33. Kxe3 {(Kf2xe3 Rg8xg3+ Ke3-f2) -2.49/6 0} Rxg3+ {(g8g3)
+4.22/10 16} 34. Kf2 {(Ke3-f2 Rg3-g4 Rh4xg4 Be6xg4 Rd1-b1 f6xe5) -2.49/6}
Rxc3 {(g3c3) +4.22/10 16} 35. exf6 {(e5xf6 Ke8-f7 Bf1-d3 Be6-b3 Bd3-g6+
Kf7xf6 Rd1-b1) -2.55/6 0} Kf7 {(e8f7) +4.05/10 16} 36. Bd3 {(Bf1-d3 Rh7-h8
Bd3-g6+ Kf7xf6 Rd1-b1 Rc3-b3 Rb1xb3 Be6xb3) -2.31/6 0} Rh6 {(h7h6) +4.09/10
16} 37. Be4 {(Bd3-e4 Kf7xf6 Rd1-b1 Rc3-b3 Rb1-e1 Rb3-c3) -2.88/6 0} Kxf6
{(f7f6) +4.09/10 15} 38. Rb1 {(Rd1-b1 Nc6-a5 Rb1-e1 Rc3-c5 Re1-g1 Be6-f5
Be4xf5) -2.66/6 0} b6 {(b7b6) +4.20/10 15} 39. Rg1 {(Rb1-g1 Rc3-c5) -3.00/6
0} Nd4 {(c6d4) +4.38/10 15} 40. Rb1 {(Rg1-b1 Nd4-b3) -3.75/6 0} Bf5 {(e6f5)
+4.80/10 15} 41. Rd1 {(Rb1-d1 Bf5xe4 Rd1xd4 Be4-g6 Rd4-a4 Bg6xh5) -3.89/6
0} c5 {(c7c5) +5.02/11 15} 42. Bd5 {(Be4-d5 Bf5-g6 Rd1-h1 Rc3-c2+ Kf2-g3
Nd4-f5+ Kg3-g4 Nf5-e3+ Kg4-f3) -4.69/6 0} Rc2+ {(c3c2) +5.48/10 14} 43. Kf1
{(Kf2-f1 Bf5-g6 Bd5-b3 Rc2-b2 Rd1xd4 c5xd4) -5.20/6 0} Bg6 {(f5g6) +5.62/10
14} 44. Bb3 {(Bd5-b3 Rc2-b2 Bb3-d5 Bg6xh5 Rd1-e1 Rb2-c2) -5.17/6 0} Rb2
{(c2b2) +7.25/11 14} 45. Bc4 {(Bb3-c4 Nd4-f5 Rh4-h3 Rh6xh5 Rh3xh5 Nf5-e3+
Kf1-e1 Ne3-g2+) -5.49/6 0} Bxh5 {(g6h5) +7.76/12 14} 46. Re1 {(Rd1-e1
Bh5-e2+ Re1xe2 Rb2-b1+ Kf1-f2 Rh6xh4 Re2-a2 Rh4-h2+ Kf2-g3 Rh2xa2) -6.72/6
0} Be2+ {(h5e2) +8.21/10 14} 47. Rxe2 {(Re1xe2 Rb2-b1+ Kf1-f2 Rh6xh4 Re2-e8
Rb1-b2+ Kf2-g3 Nd4-f5+ Kg3-f3) -6.69/6 0} Rb1+ {(b2b1) +8.58/11 14} 48. Kf2
{(Kf1-f2 Rh6xh4 Re2-e8 b6-b5 Re8-f8+ Kf6-e7 Rf8-f7+ Ke7-d6) -6.76/6 0} Rxh4
{(h6h4) +8.69/10 14} 49. Re8 {(Re2-e8 b6-b5 Re8-f8+ Kf6-g7 Rf8-f7+ Kg7-g6
f4-f5+ Kg6-g5 Bc4xb5 Nd4xb5) -7.03/6 0} Rb2+ {(b1b2) +11.67/9 13} 50. Ke3
{(Kf2-e3 b6-b5 Re8-f8+ Kf6-e7 Rf8-f7+ Ke7-d6 Bc4-f1 Rb2-b1) -6.68/6 0} Nf5+
{(d4f5) +12.70/10 13} 51. Kf3 {(Ke3-f3 Rb2-c2 Re8-e6+ Kf6-f7 Re6-e5+ Kf7-g6
Bc4-d5 Rc2-b2) -6.35/6 0} Rh3+ {(h4h3) +13.73/9 13} 52. Ke4 {(Kf3-e4
Rh3-e3+ Ke4-d5 Re3xe8 Kd5-c6 Re8-e4 Bc4-a6) -11.64/6 0} Re3+ {(h3e3)
+15.50/10 13} 53. Kd5 {(Ke4-d5 Re3xe8 Kd5-c6 Re8-e4 Bc4-a6) -11.64/5} Rxe8
{(e3e8) +15.09/9 13} 54. Ba6 {(Bc4-a6 Rb2-b4 Kd5-c6 Rb4-e4 Ba6-d3 Re4-e1
Bd3xf5) -11.63/6 0} Re6 {(e8e6) +15.50/8 13} 55. Kc4 {(Kd5-c4 Rb2-b4+
Kc4-c3 Re6-e3+ Kc3-c2 Re3-h3 Ba6-d3 Rb4xf4) -11.96/6 0} Ne3+ {(f5e3) +M5/7
0} 56. Kc3 {(Kc4-c3 Rb2-b4 Ba6-b7 Re6-d6 Bb7-a6 Ne3-d1+ Kc3-c2) -11.54/6 0
White resigns} *[/pgn]

You quote "ratings" for various engines, which I'm guessing are CCRL ratings (Rapid or Blitz?). But the ratings for Dragon levels are supposed to be FIDE scale (at Rapid), which is not even close to CCRL ratings at anything close to 2000. Your tests don't tell us much about the Komodo skill levels in human terms, unless you use some conversion formula or chart (such as ones I have posted in the past) to convert.

Yes those are from CCRL, so how do we convert them into FIDE ? http://ccrl.chessdom.com/ccrl/4040/

I made a couple posts on that topic within the last year, but I can't locate them quickly. I think my conclusion was that the two scales converge somewhere in the low 3000s, but that the ccrl ratings are spread apart by something like a 5 to 4 ratio. If we assume 3200 is the convergence point, that would mean that you add 20% of the difference between 3200 and the CCRL rating to get an estimated FIDE rating (for 15' + 10" with humans, with the engine running on the reference i7 hardware). So for example if that is correct then 1850 ccrl would mean 1850 + (0.2 x (3200 - 1850) = 1850 + 270 = 2120 FIDE. This is probably not quite right but should be at least reasonably close to correct.

I also found this article in relation from CCRL to FIDE ==> https://dokumen.tips/documents/the-leve ... tml?page=1

[Graham Banks]

I also found this article in relation from CCRL to FIDE[/size] ==> https://dokumen.tips/documents/the-leve ... tml?page=1

But it nevertheless turns out that expectations that top engines on up-to-date desktop machines are supposed to perform 3100-3200 against humans are a myth.

I think that this either shows that the author of the article has little real clue, or that the data is too outdated.

one must take into consideration that CCRL games are run on a relatively outdated hardware

A common misconception.

[Fritz 0]

After 1300 and something games, Komodo 12 level 17 has +26 Elo against Dragon 2.6 Elo 2000. This is a bit stronger than I predicted. Taking into account that now I am somewhat better than Komodo 12 level 17 at 15+10, and that I am only 1900 at chess.com rapid, there is some confusion. I estimated that Komodo 12 level 17 was about 1800, but it turns out that it's much stronger. There are only two logical answers - either I got qiute stronger than 1900 recently, or Dragon Elo 2000 is overestimated. Somehow I think that the letter is true.

There is a third plausible explanation, which I am inclined to believe. Dragon, using NNUE, is very different from Komodo, using hand-crafted eval. Dragon is dramatically stronger positionally, rarely making strategic/position errors which all HCE engines make. So for Dragon Elo 2000 to play near the level of Komodo 17 level 17 in direct matches, it has to be much weaker tactically (due to a much shorter search depth). Against humans even this shorter search is usually enough, and the vastly better positional play makes all the difference against humans, but in direct matches being outsearched by 3 plies or so is a really big deal. So even though you tend to beat Komodo 12 level 17, you might struggle against Dragon Elo 2000. But I would also say that if you can really make a plus score against Komodo 12 level 17, I think you must be stronger than 1900. I recall that level 17 was a pretty good match for average GMs in slow blitz (5' + 2"), and I can't imagine that a 2500 rated GM playing slow blitz would have much trouble beating a 1900 player playing Rapid. Do you have any other ratings you can quote, such as chess.com blitz, Fide, Lichess blitz or Rapid, etc.? I know they are not directly comparable, but I know how to convert from one scale to another reasonably well.

Larrry, I had made a typo, which I edited. I meant Komodo 12 level 14. So Komodo 12 level 14 is beating Dragon 2.6 Elo 2000 by 20+ Elo. I surely can make a solid plus score against it at 15+10 recently. I can even put a decent resistance to it at 5'+5''. My Lichess blitz rating is slightly above 2000, and rapid slightly above 2100.

[lkaufman]

After 1300 and something games, Komodo 12 level 17 has +26 Elo against Dragon 2.6 Elo 2000. This is a bit stronger than I predicted. Taking into account that now I am somewhat better than Komodo 12 level 17 at 15+10, and that I am only 1900 at chess.com rapid, there is some confusion. I estimated that Komodo 12 level 17 was about 1800, but it turns out that it's much stronger. There are only two logical answers - either I got qiute stronger than 1900 recently, or Dragon Elo 2000 is overestimated. Somehow I think that the letter is true.

There is a third plausible explanation, which I am inclined to believe. Dragon, using NNUE, is very different from Komodo, using hand-crafted eval. Dragon is dramatically stronger positionally, rarely making strategic/position errors which all HCE engines make. So for Dragon Elo 2000 to play near the level of Komodo 17 level 17 in direct matches, it has to be much weaker tactically (due to a much shorter search depth). Against humans even this shorter search is usually enough, and the vastly better positional play makes all the difference against humans, but in direct matches being outsearched by 3 plies or so is a really big deal. So even though you tend to beat Komodo 12 level 17, you might struggle against Dragon Elo 2000. But I would also say that if you can really make a plus score against Komodo 12 level 17, I think you must be stronger than 1900. I recall that level 17 was a pretty good match for average GMs in slow blitz (5' + 2"), and I can't imagine that a 2500 rated GM playing slow blitz would have much trouble beating a 1900 player playing Rapid. Do you have any other ratings you can quote, such as chess.com blitz, Fide, Lichess blitz or Rapid, etc.? I know they are not directly comparable, but I know how to convert from one scale to another reasonably well.

Larrry, I had made a typo, which I edited. I meant Komodo 12 level 14. So Komodo 12 level 14 is beating Dragon 2.6 Elo 2000 by 20+ Elo. I surely can make a solid plus score against it at 15+10 recently. I can even put a decent resistance to it at 5'+5''. My Lichess blitz rating is slightly above 2000, and rapid slightly above 2100.

OK, that makes more sense. Lichess Rapid 2100 is pretty consistent with chess.com Rapid 1900 and also with FIDE 1900, so I guess that's a fair rating for you. In your above paragraph, when you say you make a solid plus score against "it", do you mean Komodo 12 level 14 or Dragon 2.6 Elo 2000? If you are actually beating Dragon 2.6 Elo 2000 in Rapid I am concerned, as I am trying to get the Elo ratings right in Dragon 3, and Dragon 3 as currently planned will be weaker at Elo 2000 than Dragon 2.6 since I was convinced that Dragon 2.6 was too strong at elo settings anywhere near or above 2000. But if you are just projecting that you might beat it based on results vs. Komodo 12 I'm much less concerned, as explained above.

[Fritz 0]

After 1300 and something games, Komodo 12 level 17 has +26 Elo against Dragon 2.6 Elo 2000. This is a bit stronger than I predicted. Taking into account that now I am somewhat better than Komodo 12 level 17 at 15+10, and that I am only 1900 at chess.com rapid, there is some confusion. I estimated that Komodo 12 level 17 was about 1800, but it turns out that it's much stronger. There are only two logical answers - either I got qiute stronger than 1900 recently, or Dragon Elo 2000 is overestimated. Somehow I think that the letter is true.

There is a third plausible explanation, which I am inclined to believe. Dragon, using NNUE, is very different from Komodo, using hand-crafted eval. Dragon is dramatically stronger positionally, rarely making strategic/position errors which all HCE engines make. So for Dragon Elo 2000 to play near the level of Komodo 17 level 17 in direct matches, it has to be much weaker tactically (due to a much shorter search depth). Against humans even this shorter search is usually enough, and the vastly better positional play makes all the difference against humans, but in direct matches being outsearched by 3 plies or so is a really big deal. So even though you tend to beat Komodo 12 level 17, you might struggle against Dragon Elo 2000. But I would also say that if you can really make a plus score against Komodo 12 level 17, I think you must be stronger than 1900. I recall that level 17 was a pretty good match for average GMs in slow blitz (5' + 2"), and I can't imagine that a 2500 rated GM playing slow blitz would have much trouble beating a 1900 player playing Rapid. Do you have any other ratings you can quote, such as chess.com blitz, Fide, Lichess blitz or Rapid, etc.? I know they are not directly comparable, but I know how to convert from one scale to another reasonably well.

Larrry, I had made a typo, which I edited. I meant Komodo 12 level 14. So Komodo 12 level 14 is beating Dragon 2.6 Elo 2000 by 20+ Elo. I surely can make a solid plus score against it at 15+10 recently. I can even put a decent resistance to it at 5'+5''. My Lichess blitz rating is slightly above 2000, and rapid slightly above 2100.

OK, that makes more sense. Lichess Rapid 2100 is pretty consistent with chess.com Rapid 1900 and also with FIDE 1900, so I guess that's a fair rating for you. In your above paragraph, when you say you make a solid plus score against "it", do you mean Komodo 12 level 14 or Dragon 2.6 Elo 2000? If you are actually beating Dragon 2.6 Elo 2000 in Rapid I am concerned, as I am trying to get the Elo ratings right in Dragon 3, and Dragon 3 as currently planned will be weaker at Elo 2000 than Dragon 2.6 since I was convinced that Dragon 2.6 was too strong at elo settings anywhere near or above 2000. But if you are just projecting that you might beat it based on results vs. Komodo 12 I'm much less concerned, as explained above.

Actually, I have played only two games against Dragon 2.6 Elo 2000 at 15+10, and I won both. I've played much more against Komodo 12 level 14 and I have a clear plus, but don't know the exact result. I've noticed that Dragon 2.6 Elo 2000 made unexpected tactical mistakes, or maybe speculative unsound sacrifices, which Komodo 12 level 14 didn't.

[lkaufman]

After 1300 and something games, Komodo 12 level 17 has +26 Elo against Dragon 2.6 Elo 2000. This is a bit stronger than I predicted. Taking into account that now I am somewhat better than Komodo 12 level 17 at 15+10, and that I am only 1900 at chess.com rapid, there is some confusion. I estimated that Komodo 12 level 17 was about 1800, but it turns out that it's much stronger. There are only two logical answers - either I got qiute stronger than 1900 recently, or Dragon Elo 2000 is overestimated. Somehow I think that the letter is true.

There is a third plausible explanation, which I am inclined to believe. Dragon, using NNUE, is very different from Komodo, using hand-crafted eval. Dragon is dramatically stronger positionally, rarely making strategic/position errors which all HCE engines make. So for Dragon Elo 2000 to play near the level of Komodo 17 level 17 in direct matches, it has to be much weaker tactically (due to a much shorter search depth). Against humans even this shorter search is usually enough, and the vastly better positional play makes all the difference against humans, but in direct matches being outsearched by 3 plies or so is a really big deal. So even though you tend to beat Komodo 12 level 17, you might struggle against Dragon Elo 2000. But I would also say that if you can really make a plus score against Komodo 12 level 17, I think you must be stronger than 1900. I recall that level 17 was a pretty good match for average GMs in slow blitz (5' + 2"), and I can't imagine that a 2500 rated GM playing slow blitz would have much trouble beating a 1900 player playing Rapid. Do you have any other ratings you can quote, such as chess.com blitz, Fide, Lichess blitz or Rapid, etc.? I know they are not directly comparable, but I know how to convert from one scale to another reasonably well.

Larrry, I had made a typo, which I edited. I meant Komodo 12 level 14. So Komodo 12 level 14 is beating Dragon 2.6 Elo 2000 by 20+ Elo. I surely can make a solid plus score against it at 15+10 recently. I can even put a decent resistance to it at 5'+5''. My Lichess blitz rating is slightly above 2000, and rapid slightly above 2100.

OK, that makes more sense. Lichess Rapid 2100 is pretty consistent with chess.com Rapid 1900 and also with FIDE 1900, so I guess that's a fair rating for you. In your above paragraph, when you say you make a solid plus score against "it", do you mean Komodo 12 level 14 or Dragon 2.6 Elo 2000? If you are actually beating Dragon 2.6 Elo 2000 in Rapid I am concerned, as I am trying to get the Elo ratings right in Dragon 3, and Dragon 3 as currently planned will be weaker at Elo 2000 than Dragon 2.6 since I was convinced that Dragon 2.6 was too strong at elo settings anywhere near or above 2000. But if you are just projecting that you might beat it based on results vs. Komodo 12 I'm much less concerned, as explained above.

Actually, I have played only two games against Dragon 2.6 Elo 2000 at 15+10, and I won both. I've played much more against Komodo 12 level 14 and I have a clear plus, but don't know the exact result. I've noticed that Dragon 2.6 Elo 2000 made unexpected tactical mistakes, or maybe speculative unsound sacrifices, which Komodo 12 level 14 didn't.

OK, well we did fix one particular type of blunder seen often with Elo settings around 2000 or so in Dragon 2.6 (or 2.6.1), basically blindness to "sham" sacrifices (moves that at first glance look like blunders but actually win material or mate). So while Dragon 3 at Elo 2000 will tend to lose to Dragon 2.6 at same Elo, it won't make such obvious blunders and so may be a better opponent for a human.

[Chessqueen]

After 1300 and something games, Komodo 12 level 17 has +26 Elo against Dragon 2.6 Elo 2000. This is a bit stronger than I predicted. Taking into account that now I am somewhat better than Komodo 12 level 17 at 15+10, and that I am only 1900 at chess.com rapid, there is some confusion. I estimated that Komodo 12 level 17 was about 1800, but it turns out that it's much stronger. There are only two logical answers - either I got qiute stronger than 1900 recently, or Dragon Elo 2000 is overestimated. Somehow I think that the letter is true.

There is a third plausible explanation, which I am inclined to believe. Dragon, using NNUE, is very different from Komodo, using hand-crafted eval. Dragon is dramatically stronger positionally, rarely making strategic/position errors which all HCE engines make. So for Dragon Elo 2000 to play near the level of Komodo 17 level 17 in direct matches, it has to be much weaker tactically (due to a much shorter search depth). Against humans even this shorter search is usually enough, and the vastly better positional play makes all the difference against humans, but in direct matches being outsearched by 3 plies or so is a really big deal. So even though you tend to beat Komodo 12 level 17, you might struggle against Dragon Elo 2000. But I would also say that if you can really make a plus score against Komodo 12 level 17, I think you must be stronger than 1900. I recall that level 17 was a pretty good match for average GMs in slow blitz (5' + 2"), and I can't imagine that a 2500 rated GM playing slow blitz would have much trouble beating a 1900 player playing Rapid. Do you have any other ratings you can quote, such as chess.com blitz, Fide, Lichess blitz or Rapid, etc.? I know they are not directly comparable, but I know how to convert from one scale to another reasonably well.

Larrry, I had made a typo, which I edited. I meant Komodo 12 level 14. So Komodo 12 level 14 is beating Dragon 2.6 Elo 2000 by 20+ Elo. I surely can make a solid plus score against it at 15+10 recently. I can even put a decent resistance to it at 5'+5''. My Lichess blitz rating is slightly above 2000, and rapid slightly above 2100.

OK, that makes more sense. Lichess Rapid 2100 is pretty consistent with chess.com Rapid 1900 and also with FIDE 1900, so I guess that's a fair rating for you. In your above paragraph, when you say you make a solid plus score against "it", do you mean Komodo 12 level 14 or Dragon 2.6 Elo 2000? If you are actually beating Dragon 2.6 Elo 2000 in Rapid I am concerned, as I am trying to get the Elo ratings right in Dragon 3, and Dragon 3 as currently planned will be weaker at Elo 2000 than Dragon 2.6 since I was convinced that Dragon 2.6 was too strong at elo settings anywhere near or above 2000. But if you are just projecting that you might beat it based on results vs. Komodo 12 I'm much less concerned, as explained above.

Actually, I have played only two games against Dragon 2.6 Elo 2000 at 15+10, and I won both. I've played much more against Komodo 12 level 14 and I have a clear plus, but don't know the exact result. I've noticed that Dragon 2.6 Elo 2000 made unexpected tactical mistakes, or maybe speculative unsound sacrifices, which Komodo 12 level 14 didn't.

OK, well we did fix one particular type of blunder seen often with Elo settings around 2000 or so in Dragon 2.6 (or 2.6.1), basically blindness to "sham" sacrifices (moves that at first glance look like blunders but actually win material or mate). So while Dragon 3 at Elo 2000 will tend to lose to Dragon 2.6 at same Elo, it won't make such obvious blunders and so may be a better opponent for a human.

Are you planning to release Dragon 3.0 by the 1st week of May, and if so will Komodo 14.1 become Free or the price will be lower to only $19.99 ?

[lkaufman]

After 1300 and something games, Komodo 12 level 17 has +26 Elo against Dragon 2.6 Elo 2000. This is a bit stronger than I predicted. Taking into account that now I am somewhat better than Komodo 12 level 17 at 15+10, and that I am only 1900 at chess.com rapid, there is some confusion. I estimated that Komodo 12 level 17 was about 1800, but it turns out that it's much stronger. There are only two logical answers - either I got qiute stronger than 1900 recently, or Dragon Elo 2000 is overestimated. Somehow I think that the letter is true.

There is a third plausible explanation, which I am inclined to believe. Dragon, using NNUE, is very different from Komodo, using hand-crafted eval. Dragon is dramatically stronger positionally, rarely making strategic/position errors which all HCE engines make. So for Dragon Elo 2000 to play near the level of Komodo 17 level 17 in direct matches, it has to be much weaker tactically (due to a much shorter search depth). Against humans even this shorter search is usually enough, and the vastly better positional play makes all the difference against humans, but in direct matches being outsearched by 3 plies or so is a really big deal. So even though you tend to beat Komodo 12 level 17, you might struggle against Dragon Elo 2000. But I would also say that if you can really make a plus score against Komodo 12 level 17, I think you must be stronger than 1900. I recall that level 17 was a pretty good match for average GMs in slow blitz (5' + 2"), and I can't imagine that a 2500 rated GM playing slow blitz would have much trouble beating a 1900 player playing Rapid. Do you have any other ratings you can quote, such as chess.com blitz, Fide, Lichess blitz or Rapid, etc.? I know they are not directly comparable, but I know how to convert from one scale to another reasonably well.

Larrry, I had made a typo, which I edited. I meant Komodo 12 level 14. So Komodo 12 level 14 is beating Dragon 2.6 Elo 2000 by 20+ Elo. I surely can make a solid plus score against it at 15+10 recently. I can even put a decent resistance to it at 5'+5''. My Lichess blitz rating is slightly above 2000, and rapid slightly above 2100.

OK, that makes more sense. Lichess Rapid 2100 is pretty consistent with chess.com Rapid 1900 and also with FIDE 1900, so I guess that's a fair rating for you. In your above paragraph, when you say you make a solid plus score against "it", do you mean Komodo 12 level 14 or Dragon 2.6 Elo 2000? If you are actually beating Dragon 2.6 Elo 2000 in Rapid I am concerned, as I am trying to get the Elo ratings right in Dragon 3, and Dragon 3 as currently planned will be weaker at Elo 2000 than Dragon 2.6 since I was convinced that Dragon 2.6 was too strong at elo settings anywhere near or above 2000. But if you are just projecting that you might beat it based on results vs. Komodo 12 I'm much less concerned, as explained above.

Actually, I have played only two games against Dragon 2.6 Elo 2000 at 15+10, and I won both. I've played much more against Komodo 12 level 14 and I have a clear plus, but don't know the exact result. I've noticed that Dragon 2.6 Elo 2000 made unexpected tactical mistakes, or maybe speculative unsound sacrifices, which Komodo 12 level 14 didn't.

OK, well we did fix one particular type of blunder seen often with Elo settings around 2000 or so in Dragon 2.6 (or 2.6.1), basically blindness to "sham" sacrifices (moves that at first glance look like blunders but actually win material or mate). So while Dragon 3 at Elo 2000 will tend to lose to Dragon 2.6 at same Elo, it won't make such obvious blunders and so may be a better opponent for a human.

Are you planning to release Dragon 3.0 by the 1st week of May, and if so will Komodo 14.1 become Free or the price will be lower to only $19.99 ?

I hope so (by first week of May), but no promise. Probably we should then make Komodo 13 free (one up from current free Komodo 12), and reduce the price of Komodo 14.1, but we haven't discussed these details yet.