Moderator: Ras
Code: Select all
2026-09-22 16:20 | chessprogramming.orgCode: Select all
general knowledge, not fetched | standard algorithm ideas, no code
Curious what were the fules?
It's a good guess that only a few people have even bothered to look at many of the LLM games at more serious, human-compatible time controls (meaning a speed at which humans can actually follow the games live).Steve Maughan wrote: ↑Thu Sep 24, 2026 2:13 pm The engines seem to be somewhat different. But I would say they are play quite a boring game — especially Opus 5.5. In the past I have drooled over the play of engines like The King, Chess Tiger, Junior and even Stockfish. They all have a unique and interesting style. There's none of that with these engines.
— Steve
This is interesting. They are certainly close. It's quite a tight loop and tight logic. Do other engines implement in a completely different way?syzygy wrote: ↑Sat Oct 03, 2026 3:03 pm Impressive, but the source code shows many very close similarities to Stockfish.
For example, SF's see_ge():
https://github.com/official-stockfish/S ... 1389-L1492
Opus 5.5's see_ge():
https://github.com/stevemaughan/opus-5. ... #L371-L416
These two implementations seem to be completely identical.
Yes - a very watchable game!Peter Berger wrote: ↑Sat Oct 03, 2026 2:00 pmIt's a good guess that only a few people have even bothered to look at many of the LLM games at more serious, human-compatible time controls (meaning a speed at which humans can actually follow the games live).
I could drool over the following game. 12...Bg6 and 19...Na8 look somewhat fishy at first sight, but Stockfish says: no big deal, sir.
Other than that, this looks like a strategic masterpiece to me: the way the knights outmanoeuvre the bishops here, in a position where it is not at all obvious that the knights are even better in the first place.
(In case anyone wonders: Opus is pondering the "Fritz way" here — the interface feeds the second move of the main line to the engine and lets it think. If there is no ponder hit, it takes the move back. It looks just like normal pondering.)
Peter
[pgn]
[Event "Lang 120min+10sek"]
[Site "Berlin"]
[Date "2026.10.03"]
[Round "?"]
[White "Crafty 25.6"]
[Black "Opus 5.5 chess 24hrs"]
[Result "0-1"]
[ECO "E46"]
[PlyCount "204"]
[TimeControl "7200+10"]
{4096MB, DESKTOP-8OCGGEO} 1. d4 {[%emt 0:00:00]} Nf6 {[%eval 13,36] [%emt 0:06:55]} 2. c4 {[%emt 0:00:06]} e6 {[%eval 20,36] [%emt 0:04:17]} 3. Nc3 {[%emt 0:00:07] (Sf3)} Bb4 {[%eval 20,36] [%emt 0:04:46]} 4. e3 {[%emt 0:00:07]} O-O {[%eval 10,35] [%emt 0:02:48]} 5. Nge2 {[%emt 0:00:07] (Ld3)} d5 {[%eval 9,34] [%emt 0:02:17]} 6. a3 {[%emt 0:00:06]} Be7 {[%eval 6,36] [%emt 0:06:27]} 7. cxd5 {[%emt 0:00:08] (Sf4)} exd5 {[%eval 0,35] [%emt 0:01:56]} 8. g3 {[%emt 0:00:08] (Sg3)} Nbd7 {[%eval -28,38] [%emt 0:04:58]} 9. Bg2 {[%emt 0:00:08]} Nb6 {[%eval -28,40] [%emt 0:02:00]} 10. O-O {[%emt 0:00:06]} Re8 {[%eval -26,40] [%emt 0:06:19]} 11. b3 {[%emt 0:00:07] (Sf4)} Bf5 {[%eval -33,39] [%emt 0:02:24]} 12. f3 {[%emt 0:03:07] (a4)} Bg6 {[%eval -22,32] [%emt 0:03:12]} 13. Nf4 {[%emt 0:07:11]} c5 {[%eval 0,34] [%emt 0:00:01]} 14. Bh3 {[%emt 0:04:21] (Sce2)} a6 {[%eval -18,33] [%emt 0:03:03]} 15. Nxg6 {[%emt 0:09:28] (Sce2)} hxg6 {[%eval -35,35] [%emt 0:02:07]} 16. Ra2 {[%emt 0:16:59]} Bd6 {[%eval -24,38] [%emt 0:00:01]} 17. Rc2 {[%emt 0:01:46] (Kg2)} Qe7 {[%eval -50,35] [%emt 0:02:02]} 18. a4 {[%emt 0:00:50]} Rad8 {[%eval -32,34] [%emt 0:00:54]} 19. Kh1 {[%emt 0:01:02] (Se2)} Na8 {[%eval -12,36] [%emt 0:05:22]} 20. Ne2 {[%emt 0:02:03]} Nb6 {[%eval -21,38] [%emt 0:01:02]} 21. dxc5 {[%emt 0:00:34] (Kg2)} Bxc5 {[%eval -43,38] [%emt 0:01:17]} 22. Nd4 {[%emt 0:00:23]} Nbd7 {[%eval -22,40] [%emt 0:07:55]} 23. Bd2 {[%emt 0:00:06] (Lxd7)} Ne5 {[%eval -57,35] [%emt 0:01:05]} 24. Bc3 {[%emt 0:03:03] (Da1)} Nc6 {[%eval -63,32] [%emt 0:01:17]} 25. Re1 {[%emt 0:01:42] (Tff2)} Bxd4 {[%eval -58,38] [%emt 0:01:25]} 26. exd4 {[%emt 0:02:53]} Qa3 {[%eval -39,39] [%emt 0:00:39]} 27. Rce2 {[%emt 0:01:07]} Rxe2 {[%eval -33,37] [%emt 0:00:01]} 28. Rxe2 {[%emt 0:01:27]} b5 {[%eval -25,37] [%emt 0:01:13]} 29. axb5 {[%emt 0:01:27]} axb5 {[%eval -31,38] [%emt 0:00:01]} 30. Qc2 {[%emt 0:01:17] (Lb2)} Qd6 {[%eval -62,32] [%emt 0:01:11]} 31. Bb2 {[%emt 0:01:30]} Nh7 {[%eval -55,37] [%emt 0:00:01]} 32. Qd3 {[%emt 0:03:02] (Dc1)} Ng5 {[%eval -78,38] [%emt 0:01:06]} 33. Bf1 {[%emt 0:01:51]} b4 {[%eval -73,38] [%emt 0:00:01]} 34. Bg2 {[%emt 0:01:30]} Ra8 {[%eval -78,36] [%emt 0:00:16]} 35. Qd2 {[%emt 0:00:46] (De3)} Qd8 {[%eval -74,40] [%emt 0:01:22]} 36. Qc1 {[%emt 0:01:53]} Na5 {[%eval -74,43] [%emt 0:00:01]} 37. Qd1 {[%emt 0:02:07] (De3)} Nh7 {[%eval -72,38] [%emt 0:01:07]} 38. Bf1 {[%emt 0:00:24] (Lh3)} Nf8 {[%eval -74,36] [%emt 0:01:58]} 39. Qd3 {[%emt 0:01:43]} Qd7 {[%eval -77,41] [%emt 0:00:01]} 40. Kg1 {[%emt 0:07:09] (Tf2)} Nc6 {[%eval -74,42] [%emt 0:01:18]} 41. Rc2 {[%emt 0:00:45] (Tf2)} Qe6 {[%eval -65,40] [%emt 0:01:55]} 42. Kg2 {[%emt 0:00:06] (Te2)} Qf6 {[%eval -74,38] [%emt 0:01:42]} 43. Be2 {[%emt 0:01:06] (Kg1)} Ne7 {[%eval -105,33] [%emt 0:01:07]} 44. f4 {[%emt 0:01:13] (Kf2)} Qe6 {[%eval -119,36] [%emt 0:02:22]} 45. Bf3 {[%emt 0:00:44]} Nf5 {[%eval -129,32] [%emt 0:00:04]} 46. Re2 {[%emt 0:00:47]} Qd7 {[%eval -116,36] [%emt 0:00:03]} 47. Re5 {[%emt 0:00:28] (h4)} Ra5 {[%eval -160,33] [%emt 0:01:19]} 48. Kg1 {[%emt 0:01:42] (h4)} Qc6 {[%eval -141,32] [%emt 0:00:49]} 49. Qd2 {[%emt 0:01:14] (Te2)} Nd7 {[%eval -228,42] [%emt 0:00:42]} 50. Re2 {[%emt 0:00:07] (Te8+)} Nf6 {[%eval -228,47] [%emt 0:00:56]} 51. Qd3 {[%emt 0:00:17]} Ra2 {[%eval -228,55] [%emt 0:00:24]} 52. Rc2 {[%emt 0:00:06]} Qa8 {[%eval -228,56] [%emt 0:00:40]} 53. g4 {[%emt 0:00:32] (Te2)} Nh4 {[%eval -295,31] [%emt 0:00:53]} 54. Be2 {[%emt 0:02:30] (Ld1)} Ne4 {[%eval -271,36] [%emt 0:00:43]} 55. Bf1 {[%emt 0:00:23]} Qd8 {[%eval -324,35] [%emt 0:00:27]} 56. Bg2 {[%emt 0:00:40] (Le2)} Nxg2 {[%eval -409,32] [%emt 0:00:54]} 57. Kxg2 {[%emt 0:00:06]} Ra8 {[%eval -437,34] [%emt 0:00:50]} 58. Qf3 {[%emt 0:00:43] (Te2)} Rc8 {[%eval -490,33] [%emt 0:00:46]} 59. Rxc8 {[%emt 0:00:44] (De2)} Qxc8 {[%eval -551,36] [%emt 0:00:37]} 60. Qd1 {[%emt 0:00:09] (De2)} Qe6 {[%eval -593,32] [%emt 0:01:05]} 61. Qe2 {[%emt 0:00:21]} Qf6 {[%eval -602,34] [%emt 0:00:39]} 62. Qf3 {[%emt 0:00:06]} Qd6 {[%eval -639,31] [%emt 0:00:34]} 63. Qf1 {[%emt 0:00:24]} Nf6 {[%eval -668,32] [%emt 0:00:40]} 64. Bc1 {[%emt 0:00:51]} Qe6 {[%eval -678,32] [%emt 0:00:01]} 65. Qf3 {[%emt 0:00:35]} Qc8 {[%eval -687,30] [%emt 0:00:02]} 66. Qd1 {[%emt 0:00:23]} Qc3 {[%eval -711,29] [%emt 0:00:13]} 67. f5 {[%emt 0:00:24]} gxf5 {[%eval -725,27] [%emt 0:00:20]} 68. gxf5 {[%emt 0:00:29]} Qa1 {[%eval -749,27] [%emt 0:00:22]} 69. Qd2 {[%emt 0:00:40]} Qb1 {[%eval -843,27] [%emt 0:00:15]} 70. Qe3 {[%emt 0:00:28]} Qc2+ {[%eval -873,28] [%emt 0:00:13]} 71. Kg1 {[%emt 0:00:33]} Qd1+ {[%eval -961,29] [%emt 0:01:35]} 72. Kg2 {[%emt 0:00:38]} Qg4+ {[%eval -984,26] [%emt 0:00:06]} 73. Kh1 {[%emt 0:00:08]} Qxf5 {[%eval -1004,27] [%emt 0:00:34]} 74. Qe2 {[%emt 0:00:12] (Df4)} Qh3 {[%eval -1084,25] [%emt 0:00:41]} 75. Be3 {[%emt 0:00:20]} Ne4 {[%eval -1316,27] [%emt 0:00:56]} 76. Kg1 {[%emt 0:00:14]} Nc3 {[%eval -1361,27] [%emt 0:00:22]} 77. Qd2 {[%emt 0:00:46] (De1)} Qf3 {[%eval -1427,34] [%emt 0:00:29]} 78. Bf2 {[%emt 0:00:14] (Df2)} Ne4 {[%eval -2538,26] [%emt 0:00:35]} 79. Qe3 {[%emt 0:00:13]} Qxf2+ {[%eval -2978,25] [%emt 0:00:30]} 80. Qxf2 {[%emt 0:00:23]} Nxf2 {[%eval -3309,30] [%emt 0:00:36]} 81. Kxf2 {[%emt 0:00:06]} g5 {[%eval -3310,27] [%emt 0:00:23]} 82. Kg3 {[%emt 0:00:05] (Ke3)} Kh7 {[%eval -3331,27] [%emt 0:04:34]} 83. Kg4 {[%emt 0:00:05]} Kg6 {[%eval -4608,21] [%emt 0:00:15]} 84. Kf3 {[%emt 0:00:05] (Kg3)} Kh5 {[%eval -4608,22] [%emt 0:00:32]} 85. Kg3 {[%emt 0:00:07] (Kg2)} f5 {[%eval -4609,26] [%emt 0:01:12]} 86. Kh3 {[%emt 0:00:05] (Kf2)} f4 {[%eval -4633,28] [%emt 0:00:25]} 87. Kg2 {[%emt 0:00:05]} Kh4 {[%eval -4657,28] [%emt 0:00:19]} 88. h3 {[%emt 0:00:15] (Kf2)} f3+ {[%eval -4684,25] [%emt 0:00:44]} 89. Kxf3 {[%emt 0:00:12] (Kh2)} Kxh3 {[%eval -32733,41] [%emt 0:00:19]} 90. Ke3 {[%emt 0:00:06] (Kf2)} g4 {[%eval -32739,40] [%emt 0:00:21]} 91. Kf4 {[%emt 0:00:05]} g3 {[%eval -32741,41] [%emt 0:00:17]} 92. Ke5 {[%emt 0:00:06]} g2 {[%eval -32743,41] [%emt 0:00:21]} 93. Kxd5 {[%emt 0:00:06] (Kd6)} g1=Q {[%eval -32745,39] [%emt 0:00:20]} 94. Kc5 {[%emt 0:00:07]} Kg4 {[%eval -32747,41] [%emt 0:00:38]} 95. Kc4 {[%emt 0:00:11]} Kf5 {[%eval -32749,45] [%emt 0:00:29]} 96. d5 {[%emt 0:00:12] (Kxb4)} Ke4 {[%eval -32751,57] [%emt 0:00:40]} 97. d6 {[%emt 0:00:11] (Kxb4)} Qd4+ {[%eval -32753,73] [%emt 0:00:43]} 98. Kb5 {[%emt 0:00:06]} Qxd6 {[%eval -32755,81] [%emt 0:00:34]} 99. Ka4 {[%emt 0:00:06] (Kc4)} Kd5 {[%eval -32759,124] [%emt 0:00:00]} 100. Kb5 {[%emt 0:00:05]} Qb8+ {[%eval -32761,124] [%emt 0:00:00]} 101. Ka4 {[%emt 0:00:05] (Ka6)} Kc6 {[%eval -32763,124] [%emt 0:00:00]} 102. Ka5 {[%emt 0:00:05]} Qb5# {[%eval -32765,124] [%emt 0:00:00]} 0-1[/pgn]
No, I just looked at the source and found similarities all over the place. Here it even "copied" the comment for captures by the king. This is just one example. It also seems to use the same TT entry format, for example (or at least 32-byte buckets, I did not check each and every field).Steve Maughan wrote: ↑Sat Oct 03, 2026 3:32 pmThis is interesting. They are certainly close. It's quite a tight loop and tight logic. Do other engines implement in a completely different way?syzygy wrote: ↑Sat Oct 03, 2026 3:03 pm Impressive, but the source code shows many very close similarities to Stockfish.
For example, SF's see_ge():
https://github.com/official-stockfish/S ... 1389-L1492
Opus 5.5's see_ge():
https://github.com/stevemaughan/opus-5. ... #L371-L416
These two implementations seem to be completely identical.
I'm not being critical but did you use AI to find this?
— Steve
One suspects that, unless told otherwise, an AI is going to give you what it has seen most often/is ubiquitous/lowest common denominator and if you don't impose a time constraint, that thing is going to be more and more like Stockfish.syzygy wrote: ↑Sat Oct 03, 2026 6:00 pmNo, I just looked at the source and found similarities all over the place. Here it even "copied" the comment for captures by the king. This is just one example. It also seems to use the same TT entry format, for example (or at least 32-byte buckets, I did not check each and every field).Steve Maughan wrote: ↑Sat Oct 03, 2026 3:32 pmThis is interesting. They are certainly close. It's quite a tight loop and tight logic. Do other engines implement in a completely different way?syzygy wrote: ↑Sat Oct 03, 2026 3:03 pm Impressive, but the source code shows many very close similarities to Stockfish.
For example, SF's see_ge():
https://github.com/official-stockfish/S ... 1389-L1492
Opus 5.5's see_ge():
https://github.com/stevemaughan/opus-5. ... #L371-L416
These two implementations seem to be completely identical.
I'm not being critical but did you use AI to find this?
— Steve