What is Komodo Average rating at Knight odds Vs Humans

[Chessqueen]

Then we have 9 games with GM Wirig (2516 FIDE), 6 with GM Perelshteyn (2494), 6 with GM Finegold (2446), 6 with IM Toth (2377), and 2 with IM Raymond Kaufman (2255). Weighted average rating of opponents = 2450. Score was 9 wins, 11 losses, 9 draws, for minus 24 elo. So performance rating giving knight odds for the 29 Rapid (15' + 10") games is FIDE 2426. Wirig and Finegold are both very strong blitz players (relative to their FIDE ratings), and we have gradually gotten stronger and learned more about settings and book, so probably 2450 FIDE who is not a blitz specialist might be about an even opponent now for us under these conditions. We'll need some sort of breakthru to win matches from 2500+ GMs this way. But we are already avoiding defeat in a majority of the GM games, it's just that we rarely win.

But what about time handicap
"Finegold will play with standard Rapid time control (15' + 10"), while Dragon will be playing blitz (3' + 2")

I think 5x time handicap is equal to 120-140 elo at this condition.Therefore Dragon's knight handicap rating should be around 2550.

I think that you are clearly wrong
I believe that with more time for Dragon Finegold can think more on the opponent time and play better and also
5x time handicap with today engines do not give 120-140 elo even with no handicap today even against engines.

I fully expect 3'+2'' to draw against 15'+10'' in at least 90% of the games today

I also agree with you Uri that with more time any GM would have more time to think while Komodo is selecting its moves, but by the same token if we replay any of the last 4 MCTS games lets say at 5'+2' or even at T/C 10'+5' for Komodo 75% of the moves would be of equal quality or even the same moves.

Note: I will replay one of the last 4 games after move 8 at the end of the opening giving Komodo Dragon 2.6.1 a T/C of 10'+5' to see if Komodo chose many different moves, if it does I will report it and re-enter the actual move from the game, and continue with the next move and so on......

I think it may play different moves relative to the games because of different net and not because of different times.

In that case Mr. Kaufman can do a little test and see how many moves would be chosen different at lets say at 5'+2' or even at T/C 10'+5'. I still believe that 75% of the moves will be exactly the same with those different T/C, and this is due to the Knight Odds

[Chessqueen]

Fourth Game
42.Na4 Na8 43.Qa7 is also a draw based on stockfish but can make the game longer.
[/quote]

I wonder why Komodo Dragon 2.6.1 played the inferior 8.Nh4 instead of 8. c3 in this position, even Stockfish 14.1 NNUE chose 8. c3? Of course I used a longer T/C of 10'+5

[pgn][Event "KnightOdds"]
[Site "Chess.com"]
[Date "2022.01.15"]
[Round "4"]
[White "PlayKomodo"]
[Black "GMBenjaminFinegold"]
[Result "*"]
[WhiteElo "3500"]
[TimeControl "600+5"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQkq - 0 1"]
[Termination "unterminated"]
[PlyCount "15"]
[WhiteType "program"]
[BlackType "human"]

1. e4 e6 2. d3 d5 3. Nd2 Nf6 4. e5 Nfd7 5. f4 c5 6. g3 Nc6 7. Nf3 f5 8. c3
{(c2-c3 a7-a5 h2-h4 a5-a4 h4-h5) -2.74/15 47} *[/pgn]

[Chessqueen]

Fourth Game
42.Na4 Na8 43.Qa7 is also a draw based on stockfish but can make the game longer.

I wonder why Komodo Dragon 2.6.1 played the inferior 8.Nh4 instead of 8. c3 in this position, even Stockfish 14.1 NNUE and Leela chose 8. c3? Of course I used a longer T/C of 10'+5

[pgn][Event "KnightOdds"]
[Site "Chess.com"]
[Date "2022.01.15"]
[Round "4"]
[White "Stockfish 14.1 NNUE"]
[Black "GMBenjaminFinegold"]
[Result "*"]
[WhiteElo "2000"]
[TimeControl "600+5"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQkq - 0 1"]
[Termination "unterminated"]
[PlyCount "15"]
[WhiteType "program"]
[BlackType "human"]

1. e4 e6 2. d3 d5 3. Nd2 Nf6 4. e5 Nfd7 5. f4 c5 6. g3 Nc6 7. Nf3 f5 8. c3
{(c2-c3) -6.53/31 44} *[/pgn]

[lkaufman]

Then we have 9 games with GM Wirig (2516 FIDE), 6 with GM Perelshteyn (2494), 6 with GM Finegold (2446), 6 with IM Toth (2377), and 2 with IM Raymond Kaufman (2255). Weighted average rating of opponents = 2450. Score was 9 wins, 11 losses, 9 draws, for minus 24 elo. So performance rating giving knight odds for the 29 Rapid (15' + 10") games is FIDE 2426. Wirig and Finegold are both very strong blitz players (relative to their FIDE ratings), and we have gradually gotten stronger and learned more about settings and book, so probably 2450 FIDE who is not a blitz specialist might be about an even opponent now for us under these conditions. We'll need some sort of breakthru to win matches from 2500+ GMs this way. But we are already avoiding defeat in a majority of the GM games, it's just that we rarely win.

But what about time handicap
"Finegold will play with standard Rapid time control (15' + 10"), while Dragon will be playing blitz (3' + 2")

I think 5x time handicap is equal to 120-140 elo at this condition.Therefore Dragon's knight handicap rating should be around 2550.

It isn't very helpful to look deeper in a lost position beyond some point. My tests with standard Komodo against other engines (or crippled dragon) indicate that beyond about 23 plies on one thread or 21 plies on many threads it isn't helpful to look deeper giving knight odds. With MCTS, it does appear that going deeper does provide some benefit when giving knight odds. So for normal mode, it is definitely not wise to give Dragon more than 3' + 2" on good hardware when giving knight odds. In MCTS mode (used in last four games vs. Finegold), it does play a bit better with more time, but giving the human more ponder time probably roughly compensates. It is possible that 6' + 4" for Dragon (for example) might have done better with MCTS than 3' + 2", but given the extra ponder time just marginally so. In short, the time odds are a very minor factor in a knight odds match.

[Chessqueen]

Then we have 9 games with GM Wirig (2516 FIDE), 6 with GM Perelshteyn (2494), 6 with GM Finegold (2446), 6 with IM Toth (2377), and 2 with IM Raymond Kaufman (2255). Weighted average rating of opponents = 2450. Score was 9 wins, 11 losses, 9 draws, for minus 24 elo. So performance rating giving knight odds for the 29 Rapid (15' + 10") games is FIDE 2426. Wirig and Finegold are both very strong blitz players (relative to their FIDE ratings), and we have gradually gotten stronger and learned more about settings and book, so probably 2450 FIDE who is not a blitz specialist might be about an even opponent now for us under these conditions. We'll need some sort of breakthru to win matches from 2500+ GMs this way. But we are already avoiding defeat in a majority of the GM games, it's just that we rarely win.

But what about time handicap
"Finegold will play with standard Rapid time control (15' + 10"), while Dragon will be playing blitz (3' + 2")

I think 5x time handicap is equal to 120-140 elo at this condition.Therefore Dragon's knight handicap rating should be around 2550.

It isn't very helpful to look deeper in a lost position beyond some point. My tests with standard Komodo against other engines (or crippled dragon) indicate that beyond about 23 plies on one thread or 21 plies on many threads it isn't helpful to look deeper giving knight odds. With MCTS, it does appear that going deeper does provide some benefit when giving knight odds. So for normal mode, it is definitely not wise to give Dragon more than 3' + 2" on good hardware when giving knight odds. In MCTS mode (used in last four games vs. Finegold), it does play a bit better with more time, but giving the human more ponder time probably roughly compensates. It is possible that 6' + 4" for Dragon (for example) might have done better with MCTS than 3' + 2", but given the extra ponder time just marginally so. In short, the time odds are a very minor factor in a knight odds match.

Versus humans it is clearly around 2426, but against other engines like Dragon 4.6 rated around 2405 it is very hard for Komodo Dragon 2.6.1 to win at Knight Odds, I took the same position that it was played vs GM Finegold and let them play it out both having a T/C 15'+10'
[pgn][Event "KnightOdds"]
[Date "2022.01.15"]
[Round "4"]
[White "Dragon-2.6.1-64bit-avx2"]
[Black "Dragon 4.6"]
[Result "0-1"]
[BlackElo "2405"]
[WhiteElo "2426"]
[TimeControl "900+10"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "121"]
[WhiteType "program"]
[BlackType "program"]

1. e4 e6 2. d3 d5 3. Nd2 Nf6 4. e5 Nfd7 5. f4 c5 6. g3 Nc6 7. Nf3 f5 8. c3
Be7 9. d4 cxd4 10. cxd4 Nb6 11. b3 Bd7 12. Kf2 O-O 13. h4 Rc8 14. Rh2 Rc7
15. Be2 Qe8 16. Kg1 Qg6 17. h5 Qe8 18. Qf1 Nb4 19. Bd2 Nc2 20. Rc1 Ba3 21.
Rb1 h6 22. Rf2 Qxh5 23. Rh2 Qg6 24. Kh1 Qxg3 25. Rh3 Qg6 26. Qf2 Rfc8 27.
Qh2 Qh7 28. Rg1 Kf8 29. Ba5 Bb4 30. Bxb4+ Nxb4 31. Bd1 a5 32. a4 Na8 33.
Ng5 Qh8 34. Rhg3 Nc6 35. Qh4 Nd8 36. Qh2 Be8 37. Nf3 Nc6 38. Qh4 Re7 39.
Qh3 Nb6 40. Be2 Nb4 41. Ng5 Qg8 42. Nf3 Qf7 43. Rxg7 Qxg7 44. Rxg7 Kxg7 45.
Qg3+ Kh8 46. Qh4 Rg7 47. Ng5 Rg6 48. Bh5 hxg5 49. fxg5 Rg7 50. Bxe8+ Kg8
51. Bh5 Nd7 52. Kg1 Nf8 53. Qf4 Rh7 54. Bf3 Nd3 55. Qd2 Rc1+ 56. Bd1 Rh1+
57. Kxh1 Rxd1+ 58. Qxd1 Nf2+ 59. Kg2 Nxd1 60. Kg3 Kg7 61. Kf3 {White
resigns} *[/pgn]

[lkaufman]

Then we have 9 games with GM Wirig (2516 FIDE), 6 with GM Perelshteyn (2494), 6 with GM Finegold (2446), 6 with IM Toth (2377), and 2 with IM Raymond Kaufman (2255). Weighted average rating of opponents = 2450. Score was 9 wins, 11 losses, 9 draws, for minus 24 elo. So performance rating giving knight odds for the 29 Rapid (15' + 10") games is FIDE 2426. Wirig and Finegold are both very strong blitz players (relative to their FIDE ratings), and we have gradually gotten stronger and learned more about settings and book, so probably 2450 FIDE who is not a blitz specialist might be about an even opponent now for us under these conditions. We'll need some sort of breakthru to win matches from 2500+ GMs this way. But we are already avoiding defeat in a majority of the GM games, it's just that we rarely win.

But what about time handicap
"Finegold will play with standard Rapid time control (15' + 10"), while Dragon will be playing blitz (3' + 2")

I think 5x time handicap is equal to 120-140 elo at this condition.Therefore Dragon's knight handicap rating should be around 2550.

It isn't very helpful to look deeper in a lost position beyond some point. My tests with standard Komodo against other engines (or crippled dragon) indicate that beyond about 23 plies on one thread or 21 plies on many threads it isn't helpful to look deeper giving knight odds. With MCTS, it does appear that going deeper does provide some benefit when giving knight odds. So for normal mode, it is definitely not wise to give Dragon more than 3' + 2" on good hardware when giving knight odds. In MCTS mode (used in last four games vs. Finegold), it does play a bit better with more time, but giving the human more ponder time probably roughly compensates. It is possible that 6' + 4" for Dragon (for example) might have done better with MCTS than 3' + 2", but given the extra ponder time just marginally so. In short, the time odds are a very minor factor in a knight odds match.

Versus humans it is clearly around 2426, but against other engines like Dragon 4.6 rated around 2405 it is very hard for Komodo Dragon 2.6.1 to win at Knight Odds, I took the same position that it was played vs GM Finegold and let them play it out both having a T/C 15'+10'
[pgn][Event "KnightOdds"]
[Date "2022.01.15"]
[Round "4"]
[White "Dragon-2.6.1-64bit-avx2"]
[Black "Dragon 4.6"]
[Result "0-1"]
[BlackElo "2405"]
[WhiteElo "2426"]
[TimeControl "900+10"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "121"]
[WhiteType "program"]
[BlackType "program"]

1. e4 e6 2. d3 d5 3. Nd2 Nf6 4. e5 Nfd7 5. f4 c5 6. g3 Nc6 7. Nf3 f5 8. c3
Be7 9. d4 cxd4 10. cxd4 Nb6 11. b3 Bd7 12. Kf2 O-O 13. h4 Rc8 14. Rh2 Rc7
15. Be2 Qe8 16. Kg1 Qg6 17. h5 Qe8 18. Qf1 Nb4 19. Bd2 Nc2 20. Rc1 Ba3 21.
Rb1 h6 22. Rf2 Qxh5 23. Rh2 Qg6 24. Kh1 Qxg3 25. Rh3 Qg6 26. Qf2 Rfc8 27.
Qh2 Qh7 28. Rg1 Kf8 29. Ba5 Bb4 30. Bxb4+ Nxb4 31. Bd1 a5 32. a4 Na8 33.
Ng5 Qh8 34. Rhg3 Nc6 35. Qh4 Nd8 36. Qh2 Be8 37. Nf3 Nc6 38. Qh4 Re7 39.
Qh3 Nb6 40. Be2 Nb4 41. Ng5 Qg8 42. Nf3 Qf7 43. Rxg7 Qxg7 44. Rxg7 Kxg7 45.
Qg3+ Kh8 46. Qh4 Rg7 47. Ng5 Rg6 48. Bh5 hxg5 49. fxg5 Rg7 50. Bxe8+ Kg8
51. Bh5 Nd7 52. Kg1 Nf8 53. Qf4 Rh7 54. Bf3 Nd3 55. Qd2 Rc1+ 56. Bd1 Rh1+
57. Kxh1 Rxd1+ 58. Qxd1 Nf2+ 59. Kg2 Nxd1 60. Kg3 Kg7 61. Kf3 {White
resigns} *[/pgn]

I think this just confirms what I've been saying, that a 2400 CCRL Rapid-rated engine is much stronger (in Rapid) than a typical 2400 FIDE rated human. The scales may be about equivalent at somewhere near the 3000 level, but the engine rating lists are spread out compared to human lists, so 2400 CCRL may be like 2600 FIDE.

[Uri Blass]

Then we have 9 games with GM Wirig (2516 FIDE), 6 with GM Perelshteyn (2494), 6 with GM Finegold (2446), 6 with IM Toth (2377), and 2 with IM Raymond Kaufman (2255). Weighted average rating of opponents = 2450. Score was 9 wins, 11 losses, 9 draws, for minus 24 elo. So performance rating giving knight odds for the 29 Rapid (15' + 10") games is FIDE 2426. Wirig and Finegold are both very strong blitz players (relative to their FIDE ratings), and we have gradually gotten stronger and learned more about settings and book, so probably 2450 FIDE who is not a blitz specialist might be about an even opponent now for us under these conditions. We'll need some sort of breakthru to win matches from 2500+ GMs this way. But we are already avoiding defeat in a majority of the GM games, it's just that we rarely win.

But what about time handicap
"Finegold will play with standard Rapid time control (15' + 10"), while Dragon will be playing blitz (3' + 2")

I think 5x time handicap is equal to 120-140 elo at this condition.Therefore Dragon's knight handicap rating should be around 2550.

It isn't very helpful to look deeper in a lost position beyond some point. My tests with standard Komodo against other engines (or crippled dragon) indicate that beyond about 23 plies on one thread or 21 plies on many threads it isn't helpful to look deeper giving knight odds. With MCTS, it does appear that going deeper does provide some benefit when giving knight odds. So for normal mode, it is definitely not wise to give Dragon more than 3' + 2" on good hardware when giving knight odds. In MCTS mode (used in last four games vs. Finegold), it does play a bit better with more time, but giving the human more ponder time probably roughly compensates. It is possible that 6' + 4" for Dragon (for example) might have done better with MCTS than 3' + 2", but given the extra ponder time just marginally so. In short, the time odds are a very minor factor in a knight odds match.

Versus humans it is clearly around 2426, but against other engines like Dragon 4.6 rated around 2405 it is very hard for Komodo Dragon 2.6.1 to win at Knight Odds, I took the same position that it was played vs GM Finegold and let them play it out both having a T/C 15'+10'
[pgn][Event "KnightOdds"]
[Date "2022.01.15"]
[Round "4"]
[White "Dragon-2.6.1-64bit-avx2"]
[Black "Dragon 4.6"]
[Result "0-1"]
[BlackElo "2405"]
[WhiteElo "2426"]
[TimeControl "900+10"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "121"]
[WhiteType "program"]
[BlackType "program"]

1. e4 e6 2. d3 d5 3. Nd2 Nf6 4. e5 Nfd7 5. f4 c5 6. g3 Nc6 7. Nf3 f5 8. c3
Be7 9. d4 cxd4 10. cxd4 Nb6 11. b3 Bd7 12. Kf2 O-O 13. h4 Rc8 14. Rh2 Rc7
15. Be2 Qe8 16. Kg1 Qg6 17. h5 Qe8 18. Qf1 Nb4 19. Bd2 Nc2 20. Rc1 Ba3 21.
Rb1 h6 22. Rf2 Qxh5 23. Rh2 Qg6 24. Kh1 Qxg3 25. Rh3 Qg6 26. Qf2 Rfc8 27.
Qh2 Qh7 28. Rg1 Kf8 29. Ba5 Bb4 30. Bxb4+ Nxb4 31. Bd1 a5 32. a4 Na8 33.
Ng5 Qh8 34. Rhg3 Nc6 35. Qh4 Nd8 36. Qh2 Be8 37. Nf3 Nc6 38. Qh4 Re7 39.
Qh3 Nb6 40. Be2 Nb4 41. Ng5 Qg8 42. Nf3 Qf7 43. Rxg7 Qxg7 44. Rxg7 Kxg7 45.
Qg3+ Kh8 46. Qh4 Rg7 47. Ng5 Rg6 48. Bh5 hxg5 49. fxg5 Rg7 50. Bxe8+ Kg8
51. Bh5 Nd7 52. Kg1 Nf8 53. Qf4 Rh7 54. Bf3 Nd3 55. Qd2 Rc1+ 56. Bd1 Rh1+
57. Kxh1 Rxd1+ 58. Qxd1 Nf2+ 59. Kg2 Nxd1 60. Kg3 Kg7 61. Kf3 {White
resigns} *[/pgn]

I think this just confirms what I've been saying, that a 2400 CCRL Rapid-rated engine is much stronger (in Rapid) than a typical 2400 FIDE rated human. The scales may be about equivalent at somewhere near the 3000 level, but the engine rating lists are spread out compared to human lists, so 2400 CCRL may be like 2600 FIDE.

I expect Dragon can do better against engines relative to humans at the same strength with knight odds because engines do not know that they need to simplify the position.
I am not sure if Jorge used the right contempt for the game and you also cannot decide based on a single game.

[lkaufman]

Then we have 9 games with GM Wirig (2516 FIDE), 6 with GM Perelshteyn (2494), 6 with GM Finegold (2446), 6 with IM Toth (2377), and 2 with IM Raymond Kaufman (2255). Weighted average rating of opponents = 2450. Score was 9 wins, 11 losses, 9 draws, for minus 24 elo. So performance rating giving knight odds for the 29 Rapid (15' + 10") games is FIDE 2426. Wirig and Finegold are both very strong blitz players (relative to their FIDE ratings), and we have gradually gotten stronger and learned more about settings and book, so probably 2450 FIDE who is not a blitz specialist might be about an even opponent now for us under these conditions. We'll need some sort of breakthru to win matches from 2500+ GMs this way. But we are already avoiding defeat in a majority of the GM games, it's just that we rarely win.

But what about time handicap
"Finegold will play with standard Rapid time control (15' + 10"), while Dragon will be playing blitz (3' + 2")

I think 5x time handicap is equal to 120-140 elo at this condition.Therefore Dragon's knight handicap rating should be around 2550.

It isn't very helpful to look deeper in a lost position beyond some point. My tests with standard Komodo against other engines (or crippled dragon) indicate that beyond about 23 plies on one thread or 21 plies on many threads it isn't helpful to look deeper giving knight odds. With MCTS, it does appear that going deeper does provide some benefit when giving knight odds. So for normal mode, it is definitely not wise to give Dragon more than 3' + 2" on good hardware when giving knight odds. In MCTS mode (used in last four games vs. Finegold), it does play a bit better with more time, but giving the human more ponder time probably roughly compensates. It is possible that 6' + 4" for Dragon (for example) might have done better with MCTS than 3' + 2", but given the extra ponder time just marginally so. In short, the time odds are a very minor factor in a knight odds match.

Versus humans it is clearly around 2426, but against other engines like Dragon 4.6 rated around 2405 it is very hard for Komodo Dragon 2.6.1 to win at Knight Odds, I took the same position that it was played vs GM Finegold and let them play it out both having a T/C 15'+10'
[pgn][Event "KnightOdds"]
[Date "2022.01.15"]
[Round "4"]
[White "Dragon-2.6.1-64bit-avx2"]
[Black "Dragon 4.6"]
[Result "0-1"]
[BlackElo "2405"]
[WhiteElo "2426"]
[TimeControl "900+10"]
[SetUp "1"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQkq - 0 1"]
[Termination "normal"]
[PlyCount "121"]
[WhiteType "program"]
[BlackType "program"]

1. e4 e6 2. d3 d5 3. Nd2 Nf6 4. e5 Nfd7 5. f4 c5 6. g3 Nc6 7. Nf3 f5 8. c3
Be7 9. d4 cxd4 10. cxd4 Nb6 11. b3 Bd7 12. Kf2 O-O 13. h4 Rc8 14. Rh2 Rc7
15. Be2 Qe8 16. Kg1 Qg6 17. h5 Qe8 18. Qf1 Nb4 19. Bd2 Nc2 20. Rc1 Ba3 21.
Rb1 h6 22. Rf2 Qxh5 23. Rh2 Qg6 24. Kh1 Qxg3 25. Rh3 Qg6 26. Qf2 Rfc8 27.
Qh2 Qh7 28. Rg1 Kf8 29. Ba5 Bb4 30. Bxb4+ Nxb4 31. Bd1 a5 32. a4 Na8 33.
Ng5 Qh8 34. Rhg3 Nc6 35. Qh4 Nd8 36. Qh2 Be8 37. Nf3 Nc6 38. Qh4 Re7 39.
Qh3 Nb6 40. Be2 Nb4 41. Ng5 Qg8 42. Nf3 Qf7 43. Rxg7 Qxg7 44. Rxg7 Kxg7 45.
Qg3+ Kh8 46. Qh4 Rg7 47. Ng5 Rg6 48. Bh5 hxg5 49. fxg5 Rg7 50. Bxe8+ Kg8
51. Bh5 Nd7 52. Kg1 Nf8 53. Qf4 Rh7 54. Bf3 Nd3 55. Qd2 Rc1+ 56. Bd1 Rh1+
57. Kxh1 Rxd1+ 58. Qxd1 Nf2+ 59. Kg2 Nxd1 60. Kg3 Kg7 61. Kf3 {White
resigns} *[/pgn]

I think this just confirms what I've been saying, that a 2400 CCRL Rapid-rated engine is much stronger (in Rapid) than a typical 2400 FIDE rated human. The scales may be about equivalent at somewhere near the 3000 level, but the engine rating lists are spread out compared to human lists, so 2400 CCRL may be like 2600 FIDE.

I expect Dragon can do better against engines relative to humans at the same strength with knight odds because engines do not know that they need to simplify the position.
I am not sure if Jorge used the right contempt for the game and you also cannot decide based on a single game.

Yes, this is all true, though some engines do understand the benefit of simplifying when ahead. I think that if he used an engine with a CCRL Rapid rating of say 2100, which is probably a roughly even match in Rapid for a human FIDE 2400, Dragon would win most of the knight odds games with any sensible settings. Also he doesn't say if it was one thread vs. one thread; in our match Komodo had 32 cores using 31 threads which more than compensates for 5 to 1 time odds. I have not yet found any engine that can score as well against Dragon at knight odds as human GMs of similar strength do, although this is dependent on my judgment of what engines are similar strength to the GMs so I could be wrong.

[Uri Blass]

[pgn][Event "banksia game"]
[Date "2022.01.19"]
[White "Dragon 2.6.1 by Komodo Chess 64-bit contempt100"]
[Black "Dragon 4.6"]
[Result "1-0"]
[TimeControl "0/900+10"]
[Time "08:40:50"]
[Termination "mate"]
[FEN "rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKB1R w KQkq - 0 1"]
[SetUp "1"]

1. c4 {-2.99/30 50062 348446611} e5 {-3.95/13 41910 100272167} 2. Nc3 {-3.20/30 49367 323442669} Nc6 {-3.94/12 38726 93061366}
3. g3 {-2.77/28 13103 91921977} Bc5 {-4.01/12 39541 94620468} 4. Bg2 {-2.86/25 13718 93120635} a5 {-4.01/12 35017 82315672}
5. O-O {-2.65/29 37297 238339723} Nge7 {-4.01/13 35776 91639087} 6. h3 {-2.83/28 39271 255798421} d6 {-4.01/12 34660 85578770}
7. d3 {-2.92/28 22041 147397976} Be6 {-4.17/11 33956 82765755} 8. Kh2 {-2.99/27 33548 218283210} Qc8 {-4.17/11 32440 77847095}
9. a3 {-2.62/25 13523 100418737} Ra6 {-4.11/10 34159 82487626} 10. e3 {-2.42/26 21484 155386128} O-O {-4.14/10 32624 78210979}
11. b3 {-2.50/27 30291 220355740} Nd8 {-4.05/11 32279 83458710} 12. d4 {-2.13/28 43211 308948038} exd4 {-4.10/12 30176 80822804}
13. exd4 {-1.62/29 17075 129232772} Bb6 {-4.10/12 28187 72599901} 14. Bb2 {-1.89/29 53291 381675909} Qd7 {-4.04/12 28701 73046769}
15. d5 {-0.58/31 22881 165081348} Bf5 {-4.01/13 27114 70439303} 16. g4 {-0.68/33 71260 482177896} Bg6 {-4.10/13 27744 71372351}
17. f4 {-0.82/29 27027 191365607} f5 {-4.25/11 27556 72449379} 18. g5 {-0.91/28 20418 149795825} Ba7 {-4.10/12 29324 85434385}
19. Qd2 {+0.00/34 75855 494193791} Bh5 {-4.03/12 26089 75051264} 20. Rae1 {+0.00/39 14500 108821670} c6 {-3.93/11 24317 66049171}
21. Ne2 {+1.20/29 14239 106104769} cxd5 {-4.02/11 24678 64628143} 22. Ng3 {+2.33/29 17414 127035358} Bg6 {-3.74/12 23573 62355717}
23. Rxe7 {+3.16/31 20115 153221718} Qxe7 {-3.22/13 43583 118357073} 24. Re1 {+3.42/35 97159 652883268} Qd7 {-3.68/13 79466 222065791}
25. Bxd5+ {+3.53/33 14810 131266506} Kh8 {-3.68/12 19861 54475922} 26. h4 {+4.48/31 14635 122246201} Re8 {-2.46/11 28303 73187221}
27. h5 {+6.15/27 14331 115481810} Ne6 {-0.93/11 23238 58414066} 28. hxg6 {+8.63/27 12997 103279338} hxg6 {+0.27/12 40257 101885554}
29. Nxf5 {+11.55/25 15820 127749355} gxf5 {+0.00/11 18320 46853766} 30. g6 {+12.52/27 11877 110419763} a4 {+0.59/10 24813 62231996}
31. Qe2 {M+14/24 9348 88487171} Kg8 {+1.25/10 17998 44373522} 32. Qh5 {M+14/21 1945 19864852} Kf8 {+5.04/9 19369 56534668}
33. Qh8+ {M+10/23 2220 4479970} Ke7 34. Qh4+ {M+9/25 3172 7612740} Kf8
35. Bf6 {M+8/25 3053 7139993} Qf7 {M-8/9 84289 393231830} 36. gxf7 {M+7/24 2783 6148170} Kxf7 {M-6/9 23868 122543053}
37. Qh5+ {M+6/23 2348 5253979} g6 {M-5/9 22330 107353010} 38. Qh6 {M+5/22 2475 25268713} Rh8 {M-4/8 13520 70149025}
39. Qxh8 {M+4/30 2757 6363238} Bg1+ {M-3/9 13913 75114098} 40. Kxg1 {M+3/32 3379 7704733} g5 {M-2/10 13623 79664135}
41. Qg7+ {M+2/47 3000 6462138} Ke8 42. Rxe6# {M+1/99 13 4830} 1-0

[/pgn]

[Uri Blass]

Dragon2.6.1 used 7 cores with contempt 100 and won easily against Dragon4.6 with knight odds in the only game that I tried at time control of 15+10.

Of course one game proves nothing but I see no reason to believe that Dragon4.6 is better in these conditions.