Skip to content
Stork.AIstork.ai
AI Tool

KataGo Review

KataGo is an open-source self-play-trained Go engine that uses deep neural networks and advanced search algorithms to achieve superhuman play.

shipped Sep 10, 2026freemium
Domain rating40
KataGo — product screenshot

Why it matters

1First released on February 27, 2019, developed by David Wu.
2Supports board sizes from 7x7 up to 50x50 with specialized networks.
3v1.18.2, released October 20, 2025, optimized CUDA backend for Turing GPUs, making transformer models over 50% faster.
4Integrated into popular Go GUIs like Lizzie and KaTrain, and online platforms such as OGS and AI Sensei.

About KataGo

Business Model
Open Source
Platforms
Web
Target Audience
Go enthusiasts, researchers
GitHubOpen Source

Specs

API Docs

View Documentation →

GitHub

View Repository →

API Available

Yes, public API

overview

What is KataGo?

KataGo is a computer Go program tool developed by David Wu that enables Go players, researchers, and developers to play, analyze, and research the game of Go using advanced AI. It is a free and open-source program, first released on February 27, 2019, capable of defeating top-level human players through an AlphaZero-like self-play training process with numerous enhancements. KataGo utilizes Monte Carlo tree search with a convolutional neural network for position evaluation and policy guidance, and has evolved to support transformer models as of v1.17.0 (July 29, 2026). Its development is supported by a distributed training effort involving community volunteers, leading to continuous improvements in neural network architectures and performance optimizations.

features

Key Features of KataGo

KataGo provides a comprehensive set of features for Go players and AI enthusiasts, leveraging deep neural networks and advanced search algorithms for strong play and detailed analysis.

  • Open-source Go engine, allowing community contributions and transparency.
  • Supports multiple board sizes, from 7x7 to 19x19, and up to 50x50 with specific networks.
  • Predicts score and territory, offering detailed insights beyond simple win rates.
  • Allows self-play training, enabling continuous improvement of its neural networks.
  • Free to contribute GPU cycles for distributed training of neural networks.
  • Uses deep neural networks for position evaluation and policy guidance.
  • Employs advanced search algorithms, specifically Monte Carlo tree search.
  • Capable of superhuman play, consistently defeating top-level human players.
  • Advanced analysis features including move-by-move analysis, win rates, and score estimates.
  • Supports various Go rulesets and arbitrary komi values.

use cases

Who Should Use KataGo?

KataGo is designed for a diverse audience within the Go community, from competitive players to AI researchers, offering tools for play, analysis, and development.

  • Strong human Go players: For training, improving skills, and playing against a superhuman AI opponent.
  • Computer Go community members (contributors): For contributing computing resources to the distributed training of neural networks.
  • Researchers/enthusiasts: For analyzing Go games with move-by-move analysis, win rates, and score estimates.
  • Developers: For integrating KataGo into online Go websites, GUIs, or for research and development in AI for Go.
  • Users of online Go websites: As it is often the default analysis engine on platforms like OGS and AI Sensei.

how to use

How to Use KataGo

KataGo can be utilized by downloading its open-source code or pre-trained neural networks, or by accessing it through integrated platforms and GUIs. Getting started typically involves setting up the engine with a compatible interface.

  • 1Download the KataGo engine and pre-trained neural networks from the official website or GitHub repository.
  • 2Install a compatible Go GUI such as Lizzie or KaTrain, which can interface with KataGo.
  • 3Configure the GUI to point to the KataGo executable and select a desired neural network.
  • 4Begin playing against the AI, analyzing games, or reviewing positions within the chosen GUI.
  • 5For distributed training, follow instructions on the KataGo website to contribute GPU cycles.
  • 6Access KataGo's analysis features directly on online platforms like the Online Go Server (OGS) or AI Sensei.

pricing

KataGo Pricing & Plans

KataGo operates on a freemium model, meaning the core engine and its capabilities are available for free. Users can download and run the software without cost, and contribute to its development by donating computing resources for neural network training.

  • Freemium: Free access to the core KataGo engine, pre-trained neural networks, and analysis features.

Pros

  • +Achieves superhuman Go play, capable of defeating top-level human players.
  • +Offers advanced analysis features including score estimation, territory prediction, and heatmaps.
  • +Free and open-source, fostering community involvement and continuous development.
  • +Supports a wide range of board sizes (7x7 to 50x50) and various Go rulesets.
  • +Continuously updated with stronger neural networks and performance optimizations, including transformer models and new backends (ROCm, ONNX Runtime).

Cons

  • −Requires technical setup for local installation and integration with GUIs.
  • −API is not directly available, limiting programmatic access without custom wrappers.
  • −Training on user data is opt-in, which may raise privacy considerations for some users.
  • −Optimal performance often requires dedicated GPU hardware, which may not be accessible to all users.
  • −While strong, its analysis can sometimes be perceived as having 'bugs' or 'mistakes' by professional human players.

Similar Tools

KataGo vs Competitors

KataGo stands out in the competitive landscape of Go AI engines due to its advanced analysis capabilities, continuous development, and open-source nature, offering distinct advantages over other prominent tools.

1
Leela Zero↗

It is an open-source Go engine that replicates DeepMind's AlphaGo Zero algorithm, learning solely through self-play without human knowledge.

While Leela Zero is a very strong engine, KataGo is generally considered to be stronger and offers more advanced features like score estimation beyond just win/loss probabilities.

2
Minigo↗

Minigo is a minimalist, open-source Go engine modeled after AlphaGo Zero, built using TensorFlow, aiming for a simpler and more accessible implementation.

Minigo offers a more basic implementation compared to KataGo, which includes a wider array of advanced features and optimizations for stronger play and comprehensive game analysis.

See every KataGo alternative, compared→

More on Stork

Related AI Tools

Other tools in this category, matched by shared tags

IAMESSMarkAnyKeepsCounselMoneyFeedProspectio.aiAI ROI & AI Governance Platform
Visit KataGo↗
Connect
⌘
GitHubgithub.com/lightvector/KataGo/releases
💬
Discorddiscord.gg/bqkZAz3

Featured in articles

One short daily email of tools worth shipping. No drip funnel.

one email a day · unsubscribe in two clicks · no third-party tracking

For builders

This page is doing a job for someone else’s tool.

AI agents read it. Buyers land on it. It answers in eight languages and over MCP. Your tool can have one like it — live in 24 hours.

List your toolWhat you get