Edit on GitHub

Human Sparring Networks

About Human Sparring Networks

Human Sparring Networks are neural networks trained on human games, which means they play much more like real humans compared to other engines, which makes them useful for:

  • Practicing against opponents of specific skill levels
  • Analyzing games from a human perspective
  • Educational purposes and chess training
  • Casual play without being overwhelming

These networks typically range from beginner to intermediate strength (approximately 1100-2200 Elo).


Network Name Approximate Elo Download Credit Notes
Maia 1100 1100 maia-1100.pb.gz University of Toronto CSSLab Run at Nodes = 1
Maia 1200 1200 maia-1200.pb.gz University of Toronto CSSLab Run at Nodes = 1
Maia 1300 1300 maia-1300.pb.gz University of Toronto CSSLab Run at Nodes = 1
Maia 1400 1400 maia-1400.pb.gz University of Toronto CSSLab Run at Nodes = 1
Maia 1500 1500 maia-1500.pb.gz University of Toronto CSSLab Run at Nodes = 1
Maia 1600 1600 maia-1600.pb.gz University of Toronto CSSLab Run at Nodes = 1
Maia 1700 1700 maia-1700.pb.gz University of Toronto CSSLab Run at Nodes = 1
Maia 1800 1800 maia-1800.pb.gz University of Toronto CSSLab Run at Nodes = 1
Maia 1900 1900 maia-1900.pb.gz University of Toronto CSSLab Run at Nodes = 1
Maia 2200 2200 maia-2200.pb.gz @CallOn84 Run at Nodes = 1

Usage Instructions

  1. Download the network file (right click → “Save link as…”)
  2. Place it in your Lc0 networks directory (typically lc0/networks)
  3. Configure your GUI or engine to use the network with the recommended settings

Frequently Asked Questions

Q: How do these compare to just limiting nodes on a strong network?
A: Human networks make more human-like mistakes and have more consistent playing styles compared to simply weakening a strong AI.

References and Resources

Last Updated: 2025-07-25