Release : Chess System Tal EXTREME

[PHILOCHESS]

Thank you, Ed and Chris. Great work!
I have been testing Chess System Tal EXTREME in several lines that are usually considered more positional than tactical, and I must say that I really like the results.
Below, I am sharing a game obtained during those tests.
Congratulations on the release!
Robert
[pgn][Event "Test Londres 3...Bf5"]
[Site "Lima"]
[Date "2026.05.29"]
[White "Chess System Tal EXTREME"]
[Black "Dragon 4.6"]
[Result "1-0"]

1. d4 Nf6 2. Bf4 d5 3. e3 Bf5 4. c4 e6 5. Qb3 b6 6. Nc3 Bd6 7. Bxd6 Qxd6
8. Rc1 dxc4 9. Bxc4 O-O 10. Nf3 Nc6 11. Be2 Rfd8 12. O-O a6 13. Nh4 Bg4 14. Bxg4 Nxg4
15. Nf3 Nf6 16. Ng5 Ng4 17. g3 Qe7 18. Nf3 Nb4 19. h3 Nf6 20. a3 Nbd5 21. Ne5 Rd6
22. Qc4 b5 23. Nxd5 Nxd5 24. Qb3 f6 25. Nf3 Nb6 26. Rc2 Nc4 27. Qa2 Qf7 28. b3 Na5
29. Rfc1 Rd7 30. a4 b4 31. Rc5 Nb7 32. R5c4 Nd6 33. Rc6 Qh5 34. Kg2 Ra7 35. a5 Qb5
36. R6c5 Qb7 37. Qe2 Ne4 38. Rc6 Rd6 39. Rxc7 Qxc7 40. Rxc7 Rxc7 41. Qd3 Nc3 42. Nd2 Rcc6
43. Nc4 Rd5 44. e4 Rd8 45. Nb6 Nb5 46. d5 Rc3 47. Qe2 e5 48. Na4 Rc7 49. Qe3 Nd4
50. Qe1 Rb8 51. Nb6 Rc3 52. d6 Rd8 53. Qf1 Nb5 54. d7 Kf8 55. Qe2 h6 56. Qd2 Rxb3
57. Qc2 Rc3 58. Qa4 Rc6 59. Qxb4+ Kf7 60. Qb3+ Kf8 61. Qb4+ Kf7 62. Qb3+ Kf8 63. Qd5 Nd4
64. h4 g5 65. h5 g4 66. Nc8 Kg7 67. Nd6 Rxd6 68. Qxd6 Kf7 69. Qxa6 Ke7 70. Qc8 Ne6
71. a6 Rxd7 72. Qb8 Nd8 73. a7 Nc6 74. a8=Q Nxb8 75. Qxb8 Rd8 76. Qc7+ Rd7 77. Qc1 Rd8
78. Qxh6 Rg8 79. Qh7+ Kf8 80. Qf5 { Resigns. } 1-0[/pgn]

[pohl4711]

If anyone is curious I think in bullet time controls Absurd 1.0 on 1 thread would be rated approximately 2900 at the CCRL blitz rating list. I have 120 games played against Baron 3.44 x64 on 1 thread (2894 at CCRL blitz) and score is 52.92% (54 wins, 19 draws, 47 losses) in favor of Absurd 1.0 1T so Absurd is performing about 20 elo stronger than Baron 3.44 x64 1T.

And if you give Absurd 1.0 27 threads I can say that its expected elo difference over Absurd 1.0 running with just 1 thread with 99% confidence is +190 Elo ± ~90 Elo in favor of the 27 thread version in bullet chess as I have had them play 186 games against each other. If I were to guess that if CCRL were to test Absurd 1.0 on 27 threads (i know it's very unlikely) it'd probably end up rated between 3000 and 3180 with 3090 as most likely.

I don't know how 2900 CCRL blitz rating translates to human rating today but I imagine a strong super grandmaster might have a small chance at drawing a match against Absurd 1 running on 1 thread.

You cannot test any of these engines by Ed and Chris properly using 27 threads. Because in these engines, the multithreading is a bad joke.
Not only it is limited to 64 threads, but much more problematic: It is not possible to give these engines more threads than physical cores on the machine. Correct: It is possible, but has no effect...
So, testing these engines using 27 threads, your machine needs at least 27 physical cores.
I have no idea, why this is the case - no other modern multithreading-engine does this so bad. And this prevents all these engines from participating in the big engine-tournaments (TCEC and chesscom), because these engines can use only 50% of computing power (normally 1 core = 2 Threads). Which is really a shame.

[Aleks Peshkov]

[White "Chess System Tal EXTREME"]
[Black "Dragon 4.6"]
[Result "1-0"]

3301Elo vs 2382 Elo match?

[Rebel]

If anyone is curious I think in bullet time controls Absurd 1.0 on 1 thread would be rated approximately 2900 at the CCRL blitz rating list. I have 120 games played against Baron 3.44 x64 on 1 thread (2894 at CCRL blitz) and score is 52.92% (54 wins, 19 draws, 47 losses) in favor of Absurd 1.0 1T so Absurd is performing about 20 elo stronger than Baron 3.44 x64 1T.

And if you give Absurd 1.0 27 threads I can say that its expected elo difference over Absurd 1.0 running with just 1 thread with 99% confidence is +190 Elo ± ~90 Elo in favor of the 27 thread version in bullet chess as I have had them play 186 games against each other. If I were to guess that if CCRL were to test Absurd 1.0 on 27 threads (i know it's very unlikely) it'd probably end up rated between 3000 and 3180 with 3090 as most likely.

I don't know how 2900 CCRL blitz rating translates to human rating today but I imagine a strong super grandmaster might have a small chance at drawing a match against Absurd 1 running on 1 thread.

You cannot test any of these engines by Ed and Chris properly using 27 threads. Because in these engines, the multithreading is a bad joke.
Not only it is limited to 64 threads, but much more problematic: It is not possible to give these engines more threads than physical cores on the machine. Correct: It is possible, but has no effect...
So, testing these engines using 27 threads, your machine needs at least 27 physical cores.
I have no idea, why this is the case - no other modern multithreading-engine does this so bad. And this prevents all these engines from participating in the big engine-tournaments (TCEC and chesscom), because these engines can use only 50% of computing power (normally 1 core = 2 Threads). Which is really a shame.

Read the documentation on GitHub.

Is this the best you can do or shall I quote the dirt you posted elsewhere?

[PHILOCHESS]

[White "Chess System Tal EXTREME"]
[Black "Dragon 4.6"]
[Result "1-0"]

3301Elo vs 2382 Elo match?

Hi Aleks, yes, you are right to raise that point.

This was not intended as a rating test or a strength comparison between these two engines. The game I shared is only one sample from several games in which I tested Chess System Tal EXTREME against different engines, some of them with similar or even higher Elo ratings.

In this particular case, I wanted to present an interesting game that came out of those experiments, especially in lines usually considered more positional than tactical. My purpose was to illustrate the style of play and the practical results I was obtaining during testing, not to suggest that CST EXTREME is stronger than Dragon 4.6.

For a proper comparison, of course, many games under controlled conditions would be needed.

Thanks for your question, Aleks. It also gives me a chance to share a little more about the context of these tests.

[chrisw]

[White "Chess System Tal EXTREME"]
[Black "Dragon 4.6"]
[Result "1-0"]

3301Elo vs 2382 Elo match?

Hi Aleks, yes, you are right to raise that point.

This was not intended as a rating test or a strength comparison between these two engines. The game I shared is only one sample from several games in which I tested Chess System Tal EXTREME against different engines, some of them with similar or even higher Elo ratings.

In this particular case, I wanted to present an interesting game that came out of those experiments, especially in lines usually considered more positional than tactical. My purpose was to illustrate the style of play and the practical results I was obtaining during testing, not to suggest that CST EXTREME is stronger than Dragon 4.6.

For a proper comparison, of course, many games under controlled conditions would be needed.

Thanks for your question, Aleks. It also gives me a chance to share a little more about the context of these tests.

Interesting points. You are both right in different ways. It’s absolutely true that to find (several or easily) games that demonstrate the style of a style engine it is more than sensible to play against weaker opponents. Stronger or equal opponents will not make it so easy for the style program to develop the game in the “manner of the style”.

I very much liked the example game you posted, I played it out looking for flashy sacrifices or king attacks, but actually what was revealed was a positional masterpiece of manoeuvre. A great joy to watch. Not quite sure why CST extreme worked its way into this sort of a game (Ed will know better because he was the training master), but anyway, thanks.

[Rebel]

[White "Chess System Tal EXTREME"]
[Black "Dragon 4.6"]
[Result "1-0"]

3301Elo vs 2382 Elo match?

Hi Aleks, yes, you are right to raise that point.

This was not intended as a rating test or a strength comparison between these two engines. The game I shared is only one sample from several games in which I tested Chess System Tal EXTREME against different engines, some of them with similar or even higher Elo ratings.

In this particular case, I wanted to present an interesting game that came out of those experiments, especially in lines usually considered more positional than tactical. My purpose was to illustrate the style of play and the practical results I was obtaining during testing, not to suggest that CST EXTREME is stronger than Dragon 4.6.

For a proper comparison, of course, many games under controlled conditions would be needed.

Thanks for your question, Aleks. It also gives me a chance to share a little more about the context of these tests.

Interesting points. You are both right in different ways. It’s absolutely true that to find (several or easily) games that demonstrate the style of a style engine it is more than sensible to play against weaker opponents. Stronger or equal opponents will not make it so easy for the style program to develop the game in the “manner of the style”.

I very much liked the example game you posted, I played it out looking for flashy sacrifices or king attacks, but actually what was revealed was a positional masterpiece of manoeuvre. A great joy to watch. Not quite sure why CST extreme worked its way into this sort of a game (Ed will know better because he was the training master), but anyway, thanks.

It's true that against obvious weaker engines balanced style engines like Cstal-Extreme or Rebel-Extreme do not necessarily have to show its aggressiveness and can win the game in a normal positional way. This is not true for our Absurd entertainment engine which is unbalanced and will play many unsound sacrifices.

[PHILOCHESS]

[White "Chess System Tal EXTREME"]
[Black "Dragon 4.6"]
[Result "1-0"]

3301Elo vs 2382 Elo match?

Hi Aleks, yes, you are right to raise that point.

This was not intended as a rating test or a strength comparison between these two engines. The game I shared is only one sample from several games in which I tested Chess System Tal EXTREME against different engines, some of them with similar or even higher Elo ratings.

In this particular case, I wanted to present an interesting game that came out of those experiments, especially in lines usually considered more positional than tactical. My purpose was to illustrate the style of play and the practical results I was obtaining during testing, not to suggest that CST EXTREME is stronger than Dragon 4.6.

For a proper comparison, of course, many games under controlled conditions would be needed.

Thanks for your question, Aleks. It also gives me a chance to share a little more about the context of these tests.

Interesting points. You are both right in different ways. It’s absolutely true that to find (several or easily) games that demonstrate the style of a style engine it is more than sensible to play against weaker opponents. Stronger or equal opponents will not make it so easy for the style program to develop the game in the “manner of the style”.

I very much liked the example game you posted, I played it out looking for flashy sacrifices or king attacks, but actually what was revealed was a positional masterpiece of manoeuvre. A great joy to watch. Not quite sure why CST extreme worked its way into this sort of a game (Ed will know better because he was the training master), but anyway, thanks.

Thank you very much, Chris. I really appreciate your comments.

Instead of a direct sacrificial attack or a flashy king hunt, CST EXTREME produced a positional and manoeuvring game, gradually increasing the pressure.

For me, that is very valuable, because it suggests that the “Tal” character of the engine is not limited to spectacular tactics, but can also appear through active piece play, latent pressure, initiative, and long-term coordination.

I will keep testing it in different types of positions and will share more examples whenever I come across something interesting.

Congratulations again to you and Ed.

[PHILOCHESS]

[White "Chess System Tal EXTREME"]
[Black "Dragon 4.6"]
[Result "1-0"]

3301Elo vs 2382 Elo match?

Hi Aleks, yes, you are right to raise that point.

This was not intended as a rating test or a strength comparison between these two engines. The game I shared is only one sample from several games in which I tested Chess System Tal EXTREME against different engines, some of them with similar or even higher Elo ratings.

In this particular case, I wanted to present an interesting game that came out of those experiments, especially in lines usually considered more positional than tactical. My purpose was to illustrate the style of play and the practical results I was obtaining during testing, not to suggest that CST EXTREME is stronger than Dragon 4.6.

For a proper comparison, of course, many games under controlled conditions would be needed.

Thanks for your question, Aleks. It also gives me a chance to share a little more about the context of these tests.

Interesting points. You are both right in different ways. It’s absolutely true that to find (several or easily) games that demonstrate the style of a style engine it is more than sensible to play against weaker opponents. Stronger or equal opponents will not make it so easy for the style program to develop the game in the “manner of the style”.

I very much liked the example game you posted, I played it out looking for flashy sacrifices or king attacks, but actually what was revealed was a positional masterpiece of manoeuvre. A great joy to watch. Not quite sure why CST extreme worked its way into this sort of a game (Ed will know better because he was the training master), but anyway, thanks.

It's true that against obvious weaker engines balanced style engines like Cstal-Extreme or Rebel-Extreme do not necessarily have to show its aggressiveness and can win the game in a normal positional way. This is not true for our Absurd entertainment engine which is unbalanced and will play many unsound sacrifices.

Hi Ed,

Apologies for moving slightly away from the main topic of this thread, which is the release of CST EXTREME.

Since you mentioned Absurd, I thought it might be interesting to share one of the games from the tests I have been running with several engines.

In this game, Absurd 1.0 defeated Dragon 4.6. What I found most interesting is that it did not win through a wild sacrifice or a direct attack on the king. The win came in a much more positional way.

Absurd gradually shifted the play to the queenside. After the early queen exchange, White expanded with a4-a5 and later b4, creating a structure in which the queenside became the main battlefield.

The decisive element was the creation of a dangerous outside passed pawn on the a-file. After White’s rook became active and the a-pawn reached a7, Black’s pieces were tied down.

I found this quite instructive, especially in relation to your comments about style engines. Even a very aggressive or unbalanced engine may sometimes express its character not only through sacrifices, but also through the way it creates practical problems and positional pressure.

[pgn][Event "Test Londres 3...Bf5"]
[Site "Lima"]
[Date "2026.05.27"]
[White "Absurd 1.0"]
[Black "Dragon 4.6"]
[Result "1-0"]

1. d4 Nf6 2. Bf4 d5 3. e3 Bf5 4. c4 e6 5. Qb3 b6 6. Nf3 Bd6 7. Bxd6 Qxd6
8. Be2 O-O 9. O-O Nbd7 10. cxd5 exd5 11. Nc3 c6 12. a4 Rfe8 13. Qa3 Qe6 14. Ng5 Qe7
15. Qxe7 Rxe7 16. a5 Ree8 17. Rfc1 Rab8 18. b4 bxa5 19. bxa5 h6 20. Nf3 Rb4 21. Na4 Re6
22. h3 Kf8 23. Nh4 Be4 24. Ra2 Bh7 25. Nf3 Ne4 26. Ne1 Bf5 27. Nd3 Rb3 28. Kh2 Kg8
29. Bf1 h5 30. f3 Ng3 31. Ndc5 Rb1 32. Rxb1 Bxb1 33. Rf2 Nxc5 34. Nxc5 Rg6 35. Kg1 Bf5
36. Rb2 Rd6 37. Rb8+ Kh7 38. Ra8 f6 39. a6 h4 40. Rxa7 Rd8 41. Rb7 Ra8 42. a7 Bc8
43. Rc7 Nxf1 44. Kxf1 { Resigns } 1-0[/pgn]

[Rebel]

[White "Chess System Tal EXTREME"]
[Black "Dragon 4.6"]
[Result "1-0"]

3301Elo vs 2382 Elo match?

Hi Aleks, yes, you are right to raise that point.

This was not intended as a rating test or a strength comparison between these two engines. The game I shared is only one sample from several games in which I tested Chess System Tal EXTREME against different engines, some of them with similar or even higher Elo ratings.

In this particular case, I wanted to present an interesting game that came out of those experiments, especially in lines usually considered more positional than tactical. My purpose was to illustrate the style of play and the practical results I was obtaining during testing, not to suggest that CST EXTREME is stronger than Dragon 4.6.

For a proper comparison, of course, many games under controlled conditions would be needed.

Thanks for your question, Aleks. It also gives me a chance to share a little more about the context of these tests.

Interesting points. You are both right in different ways. It’s absolutely true that to find (several or easily) games that demonstrate the style of a style engine it is more than sensible to play against weaker opponents. Stronger or equal opponents will not make it so easy for the style program to develop the game in the “manner of the style”.

I very much liked the example game you posted, I played it out looking for flashy sacrifices or king attacks, but actually what was revealed was a positional masterpiece of manoeuvre. A great joy to watch. Not quite sure why CST extreme worked its way into this sort of a game (Ed will know better because he was the training master), but anyway, thanks.

It's true that against obvious weaker engines balanced style engines like Cstal-Extreme or Rebel-Extreme do not necessarily have to show its aggressiveness and can win the game in a normal positional way. This is not true for our Absurd entertainment engine which is unbalanced and will play many unsound sacrifices.

Hi Ed,

Hi in return.

Apologies for moving slightly away from the main topic of this thread, which is the release of CST EXTREME.

Since you mentioned Absurd, I thought it might be interesting to share one of the games from the tests I have been running with several engines.

In this game, Absurd 1.0 defeated Dragon 4.6. What I found most interesting is that it did not win through a wild sacrifice or a direct attack on the king. The win came in a much more positional way.

Absurd gradually shifted the play to the queenside. After the early queen exchange, White expanded with a4-a5 and later b4, creating a structure in which the queenside became the main battlefield.

The decisive element was the creation of a dangerous outside passed pawn on the a-file. After White’s rook became active and the a-pawn reached a7, Black’s pieces were tied down.

I found this quite instructive, especially in relation to your comments about style engines. Even a very aggressive or unbalanced engine may sometimes express its character not only through sacrifices, but also through the way it creates practical problems and positional pressure.

It's no big surprise against a considerable lower rated engine, because Absurd or CST-EXTREME don't need to launch their fire arsenal. Against equal and especially against stronger engines it is obliged to salt opponents with fire else it will lose.

The https://rebel7775.wixsite.com/rebel/pla ... ating-list is based on an 3400 elo pool style engines has to gain style points and Absurd being the lowest elo rated engine ( est. 3000) tops the list. Patricia-3 (est. 3200) on place 4 also profits playing 3400 elo engines. For an engine like Stockfish 17.1 playing 3400 elo engines is a disadvantage, it will win most of its game in a normal positional way against 3400 engines and win not much style points.

Nothing is perfect.