Leela Chess Zero (Lc0) is a neural network chess engine inspired by AlphaZero and developed through community-driven reinforcement learning. It evaluates chess positions using trained neural networks rather than relying primarily on manually designed evaluation rules. Monte Carlo Tree Search guides move selection by exploring promising variations. Its Universal Chess Interface compatibility allows integration with graphical chess applications and analysis tools. The engine supports GPU acceleration through CUDA, ROCm, Metal, and other computational backends. CPU execution is also available through compatible numerical libraries. Users can select different neural network weights to experiment with playing strength and evaluation behavior. Lc0 supports self-play training, engine matches, and position analysis. It runs on Windows, macOS, and Linux.
Features
- Neural network-based chess evaluation using community-trained reinforcement learning models
- Monte Carlo Tree Search for exploring candidate moves and selecting promising variations
- Universal Chess Interface compatibility with chess graphical interfaces and analysis applications
- GPU acceleration through CUDA, ROCm, Metal, ONNX Runtime, and other supported backends
- Configurable neural network weights, search parameters, and hardware execution settings
- Self-play game generation, automated engine matches, and position analysis workflows