There is no OpenAI research facility called "OpenAI Lab." The phrase most often points to one of two things: the company's research work in general, or OpenAI Gym, the open-source reinforcement learning toolkit that OpenAI released in 2016. Gym is no longer maintained by OpenAI. Its repository was archived as read-only on April 8, 2026, and development continues in a fork called Gymnasium under the nonprofit Farama Foundation.
Key takeaways
- OpenAI Gym was described in a paper submitted to arXiv on June 5, 2016 by Greg Brockman, Vicki Cheung, Ludwig Pettersson, Jonas Schneider, John Schulman, Jie Tang, and Wojciech Zaremba.
- Gym provides a standard interface between a learning algorithm and a set of simulated environments, so different algorithms can be compared on the same tasks.
- The Farama Foundation says maintenance of Gym dwindled until it was unmaintained in late 2020, and it released Gymnasium, a fork of Gym 0.26.2, as the maintained successor.
- A separate project named openai_lab, owned by an independent developer on GitHub, was built on top of Gym and is not described as an OpenAI product.
What OpenAI Gym is
The Gym paper calls it a toolkit for reinforcement learning research. Reinforcement learning trains a program, often called an agent, by letting it act in an environment and receive rewards, rather than by showing it labeled examples. Before Gym, research groups wrote their own environments and benchmarks, which made published results hard to compare. The paper describes a collection of benchmark problems that share one interface, plus a website where people could post results and compare algorithm performance.
The repository's README describes the same idea today: an open-source Python library with a standard API for communication between learning algorithms and environments, and a standard set of environments that follow that API. An environment in this sense can be a simple physics control task or a video game. The library supplies the world, and the researcher supplies the learning method.
Why the name "Lab" appears
The word turns up in two ways. The first is the loose use of "OpenAI lab" for the organization itself, which began in December 2015 as a research lab. The structure of the organization is covered on its own page on this site and is left aside here. The second is a GitHub project named openai_lab, owned by a developer using the name kengz. Its page describes an experimentation framework for reinforcement learning built on OpenAI Gym, TensorFlow, and Keras, released under the MIT license. It does not claim to be an OpenAI product, and its page now points readers to a newer version called SLM-Lab. Anyone searching for "OpenAI Lab" and landing on that repository has found a third-party tool that uses Gym, not a research division.
From OpenAI to the Farama Foundation
The Farama Foundation announced its work on October 25, 2022. It describes itself as a 501(c)(3) nonprofit in the United States, run largely by volunteers, that houses open-source reinforcement learning libraries to provide standardization, long-term maintenance, and better reproducibility. Its announcement states that OpenAI created Gym, that maintenance dwindled until Gym was unmaintained in late 2020, and that in early 2021 OpenAI gave Farama control of the repository. Farama then released Gymnasium, a fork of Gym 0.26.2, as the place where future maintenance would happen.
The Gym repository's README now says the team that maintained Gym since 2021 moved all future development to Gymnasium and that Gym will not receive further updates. GitHub shows the repository as archived by its owner on April 8, 2026. Gymnasium's own documentation calls the project a maintained fork of OpenAI's Gym library and links a migration guide for older code.
What changed for people who use it
| Question | Answer as of October 2026 |
|---|---|
| Who created Gym? | OpenAI researchers, described in a 2016 paper |
| Who maintains it now? | The Farama Foundation, through Gymnasium |
| Is the original repository active? | No, archived and read-only since April 8, 2026 |
| How is code moved over? | Farama states the usual change is replacing import gym with import gymnasium as gym, with a migration guide for API differences |
| Which projects sit beside it? | PettingZoo for multi-agent tasks, Minigrid, and others listed by Farama |
| What is openai_lab? | An independent project built on Gym, with a successor named SLM-Lab |
A checklist for older reinforcement learning tutorials
Many tutorials from 2017 through 2021 use the original package. A short review before following one avoids avoidable errors:
- Check which package the code imports. The names gym and gymnasium are different packages, and the second is the maintained one.
- Check the version noted in the tutorial. Farama states its fork starts from Gym 0.26.2.
- Read the migration guide linked from the Gymnasium documentation for differences in the API between old Gym code and Gymnasium.
- Treat any page that calls an old Gym project an OpenAI "lab" product as imprecise unless it links to an OpenAI source.
What this does not say about OpenAI today
Handing off a software library says little about the organization's research. OpenAI publishes research and technical documents on its own site, and those should be read there. This article makes no claim about current research programs, funding, or product plans, and the sources used here cover only the toolkit and its maintenance history.
FAQ
Is there an official OpenAI Lab?
No page reviewed for this article describes a facility or product by that name. The phrase usually refers to OpenAI as a research organization or to Gym-related projects.
Can OpenAI Gym still be used?
The code remains available in an archived, read-only repository, but it receives no updates. Maintained work happens in Gymnasium.
Is Gymnasium made by OpenAI?
No. It is a fork maintained under the Farama Foundation, a nonprofit organization.
What did Gym measure?
It did not measure anything by itself. It supplied environments and a common interface so that researchers could run and compare their own algorithms.
Related reading
- OpenAI's research and collaboration
- OpenAI and its stated mission
- Algorithms for the traveling salesman problem
- OpenAI in education
More in AI Tools and Industry.
Last verified: October 2026
Sources
- OpenAI Gym, Brockman et al., arXiv:1606.01540, June 5, 2016
- Announcing the Farama Foundation, Farama Foundation, October 25, 2022
- openai/gym repository README, GitHub, archived April 8, 2026
- Gymnasium documentation, Farama Foundation, 2026
- kengz/openai_lab repository, GitHub, 2026