Gymnasium is the standard API for single-agent reinforcement learning environments — the maintained successor to OpenAI Gym and the common language that RL researchers and libraries speak. Maintained by the Farama Foundation, the open-source project has more than 12,700 stars on GitHub, is released under the MIT license, and was actively developed in 2026.
Nearly every reinforcement-learning workflow needs environments to train in and a consistent interface to connect them to algorithms. Gymnasium provides exactly that: a clean Env API with well-defined observation and action spaces, stepping and reset semantics, plus a large library of reference environments — from classic control and toy text problems to Atari games, MuJoCo physics simulations and board games. Because the interface is standard, algorithms from libraries like Stable-Baselines3, CleanRL and RLlib work interchangeably against the same environments, and results stay comparable across papers and projects.
Gymnasium emerged from OpenAI’s decision to hand Gym to the community, and the Farama Foundation has kept it healthy with regular releases, strict API compliance checks and a companion ecosystem — including PettingZoo for multi-agent learning. Vectorized environments, asynchronous execution and a registry for custom environments make it equally at home in a university lab practical and in large-scale distributed training runs.
Gymnasium is completely free and open source. Install it from PyPI and run environments locally; some third-party environment suites have their own licenses, but the core library costs nothing.
Gymnasium is essential infrastructure for RL researchers, students and engineers — anyone training, benchmarking or teaching reinforcement-learning agents who wants environments and tooling that follow the community standard.