v6.0

This is the first release in a few years and includes lots of changes, so I will not really attempt to enumerate them here. It includes lots of search changes and a more efficient NNUE implementation trained on positions from the Lichess elite database and scored with a previous version of the network. Estimate from Guenther Simon and Graham Banks is that this version is around 3400 CCRL.

v5.0

This release introduces an evaluation function based on a NNUE-style architecture of input size 728, with 8 buckets of 256 hidden neurons which is chosen depending on the number of pieces left on the board. This network was trained on 125 million positions from self-play games using Weiawaga v4.0. The only new search change is the introduction of reverse futility pruning. The release also includes many minor optimizations, and is ~350 Elo stronger than Weiawaga v4.0.

v4.0

This is the first release since fully porting the engine to Rust, and so that is by far the biggest change. This release also includes multithreading for the first time via Lazy SMP, the introduction of Static Exchange Evaluation, a switch to Principal Variation Search, and a host of minor optimizations made possible by writing the engine in Rust rather than Java.