This release includes a more optimized evaluation function due to Texel Tuning and some minor adjustments.
This is the final 1.x release. I learned a lot about chess programming in the past few months and want to do some things differently, so the next (major) version will be a significant rewrite of the engine to move to a more performant, incremental, NNUE-based architecture from the ground up. In the meantime, I'll gladly patch this version if some issue arises, e.g. in CCRL competitions. See you then!
Changelog
- Tune most evaluation parameters with Texel Tuning
- Search winning captures first via SEE lookup
- Fix Chess960 castling pseudo-legality test
Estimated Strength
- Lichess: 2320 (Bullet), 2250 (Blitz), 2340 (Rapid)
- Against v1.5.1 [hash=64; threads=1]: ✅ Elo difference: 40.13 +/- 17.93