Gymnasium — Standard RL Environment API

Gymnasium — Standard RL Environment API

Gymnasium (formerly OpenAI Gym) is the community standard API for reinforcement learning environments, maintained by the Farama Foundation. It defines a universal step/reset/render interface used by virtually every RL algorithm library, enabling agents to be trained across Atari 2600 games, physics simulations (MuJoCo, Box2D), robotics tasks, and custom game environments without code changes.

Key Features

• 100+ built-in environments: Atari, MuJoCo, CartPole, LunarLander, BipedalWalker, and Toy Text
• Standard gymnasium.Env interface adopted by Stable-Baselines3, CleanRL, RLlib, and all major RL frameworks
• Wrappers for reward shaping, observation normalization, frame stacking, and time limits
• Vector environments (gymnasium.vector) for parallel data collection on multi-core machines
• Python 3.10–3.14, active maintenance and bug fixes

Quick Start

pip install gymnasium[classic-control]

import gymnasium as gym

env = gym.make("CartPole-v1", render_mode="human")
obs, info = env.reset(seed=42)
for _ in range(1000):
    action = env.action_space.sample()  # random policy
    obs, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        obs, info = env.reset()
env.close()

npx ai-supply add gymnasium-rl-environments

Curated mirror of the open-source Gymnasium (MIT). Get it from the source.