DSC UCI engines

DSC (ChessDS Chessformer) is a family of neural networks that predict human chess moves, with a focus on master and grandmaster play. It independently reimplements and extends Chessformer, the architecture used by Maia-31, with a goal of improving accuracy for elite human play. The models are trained on high-rated games and fine-tuned on classical over-the-board play, assigning a probability to every legal move from the position, ratings and game context.

For more details, see the README on GitHub.
Weights on Hugging Face.

DSC models are available as six UCI engines that vary in two dimensions:

  • Model size, with larger models having more parameters. The number in each name is the approximate parameter count in millions.
  • Game history, with standard models including some previous positions, while “FEN” models are trained only with knowledge of the current position.
UCI engine downloads for Windows, Apple silicon Mac, Intel Mac and Linux
ModelWindowsIntel / AMD 64-bitMacApple siliconMacIntelLinuxIntel / AMD 64-bit
DSC-23M249.0 MB250.8 MB283.5 MB340.4 MB
DSC-79M452.0 MB453.8 MB486.5 MB543.4 MB
DSC-150M726.8 MB728.6 MB761.3 MB818.2 MB
DSC-23M-FEN249.1 MB250.9 MB283.5 MB340.4 MB
DSC-79M-FEN452.1 MB453.9 MB486.5 MB543.4 MB
DSC-150M-FEN726.8 MB728.6 MB761.3 MB818.1 MB

Mac: after extracting, paste this in Terminal and wait for bestmove, then add the engine in your GUI. Shown for DSC-23M-FEN; for another model, change the name in both places.

cd ~/Downloads/dsc-23m-fen-0.1.0-macos-arm64 && xattr -dr com.apple.quarantine . && printf 'uci\nisready\nposition startpos\ngo\nquit\n' | ./dsc-23m-fen

Licence and citation

Weights: CC BY 4.0, attribution Albert W. Hamood / ChessDS. Inference code: GPL-3.0-or-later, including its python-chess dependency.

DSC citation: Albert W. Hamood, DSC: Human Chess Move Prediction Models, ChessDS, 2026, chessds.com.

1. Architecture reference: Daniel Monroe et al., Chessformer: A Unified Architecture for Chess Modeling, ICLR 2026.