PlentyChess is a very strong UCI chess engine using an efficiently updated neural network (NNUE) for evaluation.

The innovative threat-input network is trained on 15+ billion self-generated chess positions (both standard and fischer random chess), using self-distillation techniques for part of the training, and bullet as the training library.

PlentyChess consistently ranks within the top 3 on various chess engine rating lists:

• 3rd on SPCC, 3789 Elo
• 3rd on Ipman Chess, 3597 Elo (excluding duplicate engines)
• 3rd on CEGT, 3603 Elo
• 2nd on CCRL, 3642 Elo
• 3rd on MCERL (Mac only), 3785 Elo (excluding duplicate engines)
• 3rd on e4e6 (Chess324), 3803 Elo

(Last updated: 4th November 2025)

Additionally, it plays live in the Chess.com Computer Championship and the Top Chess Engine Championship.

A big thank you goes out to:

• My friends on furybench, the testing instance used to develop PlentyChess and other strong engines
• The bullet NNUE trainer
• noobpwnftw, Styxdoto, Viren, Jay Honnold, swedishchef and Andrew Grant for donating hardware time
• fathom, which I use to probe tablebases
• incbin, which I use to embed neural network files into my binaries

I highly recommend downloading the latest release from the downloads page.

PlentyChess implements the Universal Chess Interface (UCI) protocol which is supported by basically all chess GUIs. Examples of these include Nibbler, En Croissant and Banksia.

If you want to build a development version, the following instructions might help you.

Running make (or make -j) is the easiest way to build a PlentyChess binary called engine, with the network being downloaded and embedded automatically.

This standard build contains debug symbols and is not profiled. For the fastest builds, compile with make profile-build EXE=PlentyChess -j.

You can pass an arch argument to the Makefile to compile a binary for a specific architecture (e.g. make -j arch=bmi2). The following architecture / OS combinations are currently supported:

Building Windows binaries requires both a make installation (e.g. from MinGW), as well as an LLVM installation for the clang toolchain.

For the fastest builds (non-profiled as well as profiled), use the native LLVM distribution, instead of the one contained in MinGW/msys2.

Profiled builds do not currently work with compilers installed via MinGW/msys2.

Building an Android binary requires a linux device with NDK installed.

The Makefile will automatically use the NDK compiler when invoked with arch=android.