# opendilab/LightZero

[NeurIPS 2023 Spotlight] LightZero: A Unified Benchmark for Monte Carlo Tree Search in General Sequential Decision Scenarios (awesome MCTS)

Repository: https://github.com/opendilab/LightZero
Canonical: https://ross.abutalabs.com/products/lightzero
Homepage: https://opendilab.github.io/LightZero
Language: Python
License: Apache-2.0
License Family: permissive
Topics: alpha-beta-pruning, alphazero, atari, board-games, continuous-control, gomoku, monte-carlo-tree-search, muzero, pytorch, reinforcement-learning, tictactoe, efficientzero, sampled-muzero, mcts, mcts-algorithm, board-game, gym, self-play, stochastic-muzero, gumbel-muzero
Last push: 2026-08-25T04:44:40+00:00

## Health v2 (maintenance only)
Score: 67/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 8, longevity 100
- inputs: {"age_days": 1425, "days_push": 8, "days_rel": 511, "gap_med": null, "n_releases_24m": 1}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1636, forks 197 (observed 2026-08-28T04:05:14.623585+00:00)

## What it is
LightZero is a lightweight, open-source Python toolkit that unifies Monte Carlo Tree Search (MCTS) with deep reinforcement learning, implementing algorithms like AlphaZero, MuZero, EfficientZero, Gumbel MuZero, and Stochastic MuZero. It serves as both an algorithm library and a benchmark for evaluating MCTS-based methods across board games, Atari, and continuous control tasks.

## Use cases
- train an AlphaZero agent to play Gomoku or TicTacToe from self-play
- benchmark MuZero-style algorithms on Atari environments
- apply MCTS-based RL to continuous control problems
- compare EfficientZero, Sampled MuZero, and Gumbel MuZero implementations
- learn how Monte Carlo Tree Search combined with deep RL works
- run a unified benchmark for sequential decision-making algorithms

## When to choose
- you need ready-made, well-tested implementations of MCTS+RL algorithms like MuZero or EfficientZero
- you want a unified benchmark to compare tree-search RL methods across game and control environments
- you are doing research on MCTS-based decision-making and want a PyTorch-based, customizable toolkit

## When to avoid
- you need general-purpose RL algorithms without tree search, such as PPO or SAC alone
- you want a production game AI engine rather than a research toolkit
- your project requires non-Python or extremely low-latency inference outside PyTorch

## Facets
- artifact type: library
- maturity: active
- function: reinforcement-learning, machine-learning, simulation, benchmarking, sdk
- domain: reinforcement-learning, machine-learning
- platform: python, windows
- tags: mcts, muzero, alphazero, efficientzero, self-play, board-games, pytorch, decision-making, neurips, game-development, algorithms, research, gpu, linux, macos

## Member repositories
- opendilab/LightZero (main) score 67

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:14.623585+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:46:58.551873+00:00, confidence not recorded.
  - readme: https://github.com/opendilab/LightZero (fetched 2026-08-28T04:05:14.623585+00:00, sha 083733c4da7e)
  - homepage: https://opendilab.github.io/LightZero (fetched 2026-08-29T11:20:01.725266+00:00, sha b26b4521a26e)
  - registry_pypi: https://pypi.org/pypi/lightzero/json (fetched 2026-08-29T11:20:01.749806+00:00, sha 768101891c6e)
- Data as of 2026-08-30T08:39:29.467469+00:00.