This version introduces mix9 NNUE, which is about 20-55 elo stronger than mix8.

VLTC test result on freestyle-15:
Total/Win/Draw/Lose: 4314 / 1246 / 2190 / 878
PTNML: 47 / 390 / 994 / 600 / 126
WinRate: 54.27%
ELO: 29.46[22.02, 36.98]
LOS: 100.00
LLR: 3.06[-2.94, 2.94]

VLTC test result on freestyle-20:
Total/Win/Draw/Lose: 8888 / 4709 / 21 / 4158
PTNML: 734 / 9 / 2681 / 12 / 1008
WinRate: 53.10%
ELO: 21.23[13.95, 28.53]
LOS: 100.00
LLR: 2.96[-2.94, 2.94]

VLTC test result on standard-15:
Total/Win/Draw/Lose: 3616 / 911 / 2153 / 552
PTNML: 37 / 282 / 881 / 501 / 107
WinRate: 54.96%
ELO: 34.39[26.91, 41.98]
LOS: 100.00
LLR: 3.15[-2.94, 2.94]

VLTC test result on renju-15:
Total/Win/Draw/Lose: 2734 / 1151 / 871 / 712
PTNML: 68 / 178 / 557 / 375 / 189
WinRate: 58.03%
ELO: 55.96[44.53, 67.70]
LOS: 100.00
LLR: 3.23[-2.94, 2.94]

What's New:

• New NNUE architecture, improved Elo rating
• New AVXVNNI, AVX512VNNI, ARM64-NEON, ARM64-NEON-DOTPROD instruction set engine binary
• UI font in Yixin interface is now changeable

This version introduces the new Mix8 NNUE in replacement of the previous Mix7, and it is about 75~100elo stronger than the previous release Rapfi23 (version 0.39.01).

VLTC test result on freestyle-20:
Book: f20-100M-40k
TC: 200+2
Total/Win/Draw/Lose: 7272 / 4534 / 11 / 2727
PTNML: 319 / 2 / 2087 / 9 / 1219
WinRate: 62.42%
ELO: 88.03[77.81, 98.54]
LOS: 100.00
LLR: 10.24[-6.91, 6.91]

VLTC test result on standard-15:
Book: s15-100M-40k
TC: 200+2
Total/Win/Draw/Lose: 6924 / 2458 / 3502 / 964
PTNML: 54 / 369 / 1452 / 1203 / 384
WinRate: 60.79%
ELO: 76.06[68.17, 84.19]
LOS: 100.00
LLR: 12.16[-6.91, 6.91]

VLTC test result on renju-15:
Book: r15-100M-40k
TC: 200+2
Total/Win/Draw/Lose: 6748 / 3504 / 1712 / 1532
PTNML: 107 / 309 / 1279 / 863 / 816
WinRate: 64.61%
ELO: 104.44[93.58, 115.78]
LOS: 100.00
LLR: 13.84[-6.91, 6.91]

What's New:

• New NNUE architecture and optimized search, improved Elo rating
• New AVX512 engine binary
• Texts in the database are saved in UTF-8 encoding

Thanks to @subarashikihibi for the NNUE training and search patches; @nguyencongminh090 for the new logo in readme.

Release of the full multi-threaded version of Rapfi23 (windows-x64 version).
This package contains the engine executable and a combined Yixin-board graphics user interface.

Please see the readme inside to get usage instructions.