I uploaded Stockfish after 3 months

[Jouni]

Note, there is now stockfish-windows-x86-64-universal.zip! Works nice and nps is higher than SF18. This 20260504 version is surprisingly good in test position and mates. Classic example.

[d]3qk3/8/8/8/8/8/PPPPPPPP/RNBQKBNR w KQ - 0 1

Analysis by Stockfish dev-20260504-5095cd16:

1.e4 Qd4 2.Qf3 Qg7 3.d4 Qf7 4.Bb5+ Ke7 5.Qg3 Kf8 6.Bh6+ Ke7 7.Nc3 Qh7 8.Qc7+ Kf6 9.Qxh7 Ke6 10.Bg5 Kd6 11.Qd7#
+- (#11) Depth: 70/24 00:02:47 727mN

[bmp1974]

Yes. I too noticed this. For example, the following position use to take lot of time for Stockfish to find 1. f6. But now it finds within 30 seconds.
[d]2b1rk2/5p2/p1P5/2p2P2/2p5/7B/P7/2KR4 w - - 0 1

[peter]

Yes. I too noticed this. For example, the following position use to take lot of time for Stockfish to find 1. f6. But now it finds within 30 seconds.
[d]2b1rk2/5p2/p1P5/2p2P2/2p5/7B/P7/2KR4 w - - 0 1

Why not take the original starting position of the study?

https://www.yacpdb.org/#search/MXIzazJc ... LzEvMS8w/1

SF dev. single thread (4.3GHz) still takes some time to get the winning eval, which doesn't arise till 4.f6 (the move to be found at your position 3 moves later on started):

53: Smyslov=V - (+0440.45c1f8) 1-0, 64 1938
[d]1r3k2/5p2/p1P1b1p1/2p5/2p1PP2/8/P7/2KR1B2 w - - 0 1

Analysis by Stockfish dev-20260504-5095cd16:

Code: Select all

1.f5 gxf5 2.Lh3 Te8 3.c7 Lc8 4.exf5 f6 5.Lg2 Ke7 6.Lc6 Th8 7.Td5 Tg8 8.a3 Tg1+ 9.Kd2 
  Weiß steht besser.  
  +/- (0.73)  Tiefe: 10/22   00:00:00
...
1.f5 gxf5 2.Lh3 Te8 3.c7 Lc8 4.exf5 f6 5.Lg2 Ke7 6.Te1+ Kf8 7.Td1 
  Die Stellung ist ausgeglichen.  
  = (0.00)  Tiefe: 68/13   00:01:09  145MN, tb=1484636
1.f5 gxf5 2.Lh3 Te8 3.exf5 Lc8 4.c7 f6 5.Lg2 Ke7 6.Te1+ Kf8 7.Td1 
  Die Stellung ist ausgeglichen.  
  = (0.00)  Tiefe: 69/42   00:01:26  180MN, tb=2119771
1.f5 gxf5 2.Lh3 Te8 3.exf5 Lc8 4.c7 f6 5.Lg2 Ke7 6.Te1+ Kf8 7.Td1 
  Die Stellung ist ausgeglichen.  
  = (0.00)  Tiefe: 70/13   00:01:34  196MN, tb=2276870
...
1.f5 gxf5 2.Lh3 Te8 3.exf5 Lc8 4.c7 f6 5.Lg2 Ke7 6.Te1+ Kf8 7.Td1 
  Die Stellung ist ausgeglichen.  
  = (0.01 ++)  Tiefe: 73/40   00:01:44  217MN, tb=2478927
...
1.f5 gxf5 2.Lh3 Te8 3.exf5 Lc8 4.c7 f6 5.Lg2 Ke7 6.Te1+ Kf8 7.Td1 
  Die Stellung ist ausgeglichen.  
  = (0.12 ++)  Tiefe: 73/40   00:01:45  219MN, tb=2484629
...
1.f5 gxf5 2.Lh3 Te8 3.exf5 Lc8 4.c7 f6 5.Lg2 Ke7 6.Te1+ Kf8 7.Td1 
  Weiß steht auf Gewinn.  
  +- (3.88 ++)  Tiefe: 73/85   00:02:24  309MN, tb=4107113

Here's for comparison ShashChess:

1r3k2/5p2/p1P1b1p1/2p5/2p1PP2/8/P7/2KR1B2 w - - 0 1

Analysis by ShashChess 41.2 GoldDigger:

Code: Select all

1.f5 gxf5 2.Lh3 fxe4 3.c7 Tc8 4.Td8+ Ke7 5.Txc8 Lxh3 
  Weiß steht deutlich besser.  
  +- (1.92)  Tiefe: 10/12   00:00:00
1.f5 gxf5 2.Lh3 Te8 3.exf5 Lc8 4.c7 f6 5.Lg2 Ke7 6.Lc6 Tg8 
  Weiß steht besser.  
  +/- (0.78)  Tiefe: 11/22   00:00:00
...
1.f5 gxf5 2.Lh3 Te8 3.exf5 Lc8 4.c7 f6 5.Lf1 Ke7 6.Lxc4 Th8 7.Td5 Lb7 8.Td3 Lc8 9.Td5 
  Die Stellung ist ausgeglichen.  
  = (0.00)  Tiefe: 32/17   00:00:02  3446kN, tb=8602
1.f5 gxf5 2.Lh3 Te8 3.exf5 Lc8 4.f6 Lxh3 5.c7 Lf5 6.Td8 c3 7.a4 a5 8.Kd1 c2+ 9.Kc1 c4 10.Kb2 c1D+ 11.Kxc1 Lg4 12.Kc2 Lc8 13.Kc3 Lb7 
  Weiß steht deutlich besser.  
  +- (2.19)  Tiefe: 33/37   00:00:03  5577kN, tb=22872
...
1.f5 gxf5 2.Lh3 Tc8 3.exf5 Txc6 4.fxe6 fxe6 5.Te1 Ke7 6.Lf1 Td6 7.Kc2 Td4 8.Kc3 Kd6 9.Lxc4 e5 10.Lxa6 e4 11.Lc4 Kc7 12.Te2 Kb6 13.Tb2+ Ka7 14.Tb1 Td8 15.a4 Td4 16.a5 
  Weiß steht auf Gewinn.  
  +- (3.64 ++)  Tiefe: 35/60   00:00:10  20804kN, tb=197142

[Jouni]

Reckless shows winning score fast:

Analysis by Reckless 0.10.0-dev-7300f044:
...
1.f5 gxf5 2.Bh3 Rc8 3.exf5 Rxc6 4.fxe6 fxe6 5.Re1 Ke7 6.Kc2 Rd6 7.Bf1 a5 8.Bxc4 Kf6 9.Kc3 e5 10.Rf1+ Kg5 11.Bd3 Rd5 12.Rf7 Kg4 13.Bc2 Rd4
+- (4.59) Depth: 38/59 00:00:11 75497kN, tb=133409

Generally latest Stockfish and Reckless score very similarly in my test suites.

[Eelco de Groot]

In your first position, if that is with more than one thread there you'd expect the better results;
because Mated in X are now better protected across the threads and there is thread selction possible now with Multi PV. But I only understand 0,05 % of what they were doing there so I could be all wrong again It's possible the new Net helps here in single thread as well (and other changes in the code too) but that is sometimes random? Meaning that will differ per position. If this is single thread my explanation of course fails already.