OpenSpiel — Game Theory & RL Research Framework

OpenSpiel — Game Theory & RL Research Framework

OpenSpiel is Google DeepMind's comprehensive framework for research in reinforcement learning and game theory. It implements 70+ two-player and multi-player games — from board games (Chess, Go, Hex, Breakthrough) to card games (Poker, Bridge, Skat) to matrix games and auction mechanisms — along with a library of algorithms including CFR (Counterfactual Regret Minimization), MCTS, deep RL baselines, and Nash equilibrium solvers.

Key Features

• 70+ games: perfect/imperfect information, zero-sum and general-sum
• Algorithms: CFR, CFR+, MCCFR, MCTS, AlphaZero-style search, DQN, REINFORCE, Actor-Critic
• Nash equilibrium solvers and game theory utilities
• Python and C++ APIs — same game logic in both languages
• PyTorch and TensorFlow integration for deep RL agents
• Used in AlphaStar, AlphaZero, and Pluribus research

Quick Start

pip install open_spiel

import pyspiel

game = pyspiel.load_game("chess")
state = game.new_initial_state()
print(state)  # Starting board

# Simulate a random game
import random
while not state.is_terminal():
    actions = state.legal_actions()
    action = random.choice(actions)
    state.apply_action(action)
print("Returns:", state.returns())  # [1.0, -1.0] = white wins

npx ai-supply add open-spiel-game-theory-rl

Curated mirror of the open-source OpenSpiel (Apache-2.0). Get it from the source.