Farama-Foundation/Gymnasium

What is Gymnasium?

Gymnasium is a Python library that provides a standard API for reinforcement‑learning (RL) research. It lets you create, run, and compare RL agents against a wide collection of simulated environments – from simple classic‑control tasks to complex physics‑based or Atari games. It is a community‑maintained fork of OpenAI’s original Gym library, and the place where future updates will be released.

Why it matters

• Uniform interface – All environments expose the same reset, step, action_space, and observation_space methods, so the same learning code can be swapped across many tasks.
• Reproducibility – Environments are versioned (e.g., CartPole-v1) and the version number increments whenever a change could affect learning results.
• Broad ecosystem – Includes classic control, Box2D, toy text, MuJoCo, Atari, plus many third‑party environments that already follow the Gymnasium API.
• Community support – Documentation site, Discord chat, and a clear contribution guide make it easy for newcomers to get help and for developers to add new environments.

How to get started (code snippet from the README)

import gymnasium as gym

env = gym.make("CartPole-v1")
observation, info = env.reset(seed=42)

for _ in range(1000):
    action = env.action_space.sample()          # pick a random action
    observation, reward, terminated, truncated, info = env.step(action)
    if terminated or truncated:
        observation, info = env.reset()

env.close()

Create an environment, reset it, take actions, and handle episode termination – all in just a few lines.

Installation

pip install gymnasium                # core library only
pip install "gymnasium[atari]"       # add Atari dependencies
pip install "gymnasium[all]"         # install every optional dependency

(The optional packages are large and sometimes tricky to build, so you can install only what you need.)

Related projects you might also use

Project	What it adds
CleanRL	Simple, reference RL algorithm implementations that use the Gymnasium API.
PettingZoo	Multi‑agent extension of Gymnasium with its own set of environments.
Farama Foundation environments	Additional curated environments that follow the same API.

Getting involved

Read the CONTRIBUTING.md for guidelines. Support the project financially via the GitHub Sponsors link if you can.

---

TL;DR – Gymnasium is the go‑to library for anyone building or testing reinforcement‑learning agents, offering a clean, versioned interface to a huge suite of simulated tasks.