The bleeding edge Stockfish seems to have a tactical improvement

[Dann Corbit]

I got 8 more solutions on my 120 problem test than I got for the previous version that I compiled on 2/9/2026 .
I notice that there were some large reported Elo gains and a new network.
They also did some longer time control testing.

[peter]

Thanks for reporting, Dann.
Don't want to download more often than once a month, will give first one version appearing in March a try then, regards

[Jouni]

Stockfish 18 is better than 17.1 in all my testsuites (3x100 positions). Note, that version 18 has 10% higher nodespeed even if exe file is 50% bigger!

[peter]

Nodespeed helping with EloStatTS- time indices too of course

Code: Select all

    Program                                    Elo   +/-  Matches  Score   Av.Op.   S.Pos.   MST1    MST2   RIndex
 17 Stockfish18-6t-MuPV4                     : 3511    3   9905    51.6 %   3499   232/325    4.2s   11.6s   0.64

 25 Stockfish17.1-6t-MuPV4                   : 3496    3   9818    49.4 %   3500   227/325    5.2s   12.7s   0.55

MST1  : Mean solution time (solved positions only)
MST2  : Mean solution time (solved and unsolved positions)
RIndex: Score according to solution time ranking for each position

Those 325 with 30"/pos.and 6 threads of 16x4.3GHz CPU (5 concurrencies).

viewtopic.php?p=988828#p988828

Some more runs compared in this one list since then, 38 now, so StatTS- Elo of both of them climbed a little (4 points each).

But there's quite some time and number of patches between 17.1 and 18, guess Dann writes about latest dev. -version compared to 18 already, regards

[Jouni]

Surprisingly latest Reckless scores 10 more than SF18.

[peter]

Surprisingly latest Reckless scores 10 more than SF18.

Of course depending on suite and hardware- TC the differences differ more or less

Yet with most of the anti- engine- puzzle- suites I tried so far (e.g. with the 325:

Code: Select all

    Program                                    Elo   +/-  Matches  Score   Av.Op.   S.Pos.   MST1    MST2   RIndex

  4 Reckless0.9.0-dev-2a847427-6t-MuPV4      : 3530    3  10973    54.5 %   3499   254/325    4.2s    9.8s   0.65
  
 19 Stockfish18-6t-MuPV4                     : 3510    3  10461    51.5 %   3500   232/325    4.2s   11.6s   0.64

 30 Reckless0.9.0.dev-2a847427-6t-MuPV1      : 3484    3  10430    47.5 %   3501   213/325    4.8s   13.5s   0.49

 32 Stockfish18-6t-MuPV1                     : 3471    3  10119    45.6 %   3501   204/325    5.4s   14.5s   0.49

MST1  : Mean solution time (solved positions only)
MST2  : Mean solution time (solved and unsolved positions)
RIndex: Score according to solution time ranking for each position

), Reckless has the edge over SF18, regards

[Jouni]

And there are about 50 patches after this version. Latest 1 hour ago .

[Dann Corbit]

{snip}
But there's quite some time and number of patches between 17.1 and 18, guess Dann writes about latest dev. -version compared to 18 already, regards

Exactly, it was the version from github, on the date of my posting. The clue is in the title: "The bleeding edge Stockfish"

[Jouni]

Note: SF18 now gains normally from additional cores. Version 17.1 scored basically similar with 2 or 64 cores.

[peter]

{snip}
But there's quite some time and number of patches between 17.1 and 18, guess Dann writes about latest dev. -version compared to 18 already, regards

Exactly, it was the version from github, on the date of my posting. The clue is in the title: "The bleeding edge Stockfish"

Tried the next but one now (omiting 260218) and yes, with the 325x30" there is progress, only 2 more solved positions compared to SF18 but no difference in StatTS- Elo as for MultiPV=4, yet 7 more solutions and 12 StatTS- Elo, which is clearly out of error bar as for the two single primary- runs:

Code: Select all

    Program                                    Elo   +/-  Matches  Score   Av.Op.   S.Pos.   MST1    MST2   RIndex
 26 Stockfish-260307-6t-MuPV4                : 3507    2  13940    51.1 %   3500   234/325    4.3s   11.5s   0.62
 27 Stockfish18-6t-MuPV4                     : 3507    2  13953    51.1 %   3500   232/325    4.2s   11.6s   0.64
 
 41 Stockfish-260307-6t-MuPV1                : 3480    3  13657    47.0 %   3501   211/325    5.1s   13.8s   0.55
 
 44 Stockfish18-6t-MuPV1                     : 3468    3  13489    45.3 %   3501   204/325    5.4s   14.5s   0.49
 
MST1  : Mean solution time (solved positions only)
MST2  : Mean solution time (solved and unsolved positions)
RIndex: Score according to solution time ranking for each position