Skip to content

MLGym

An open framework and benchmark for AI research agents

Visit official site

What MLGym is

The framework is described as an open-source framework and benchmark from Meta GenAI and UCSB NLP for training and evaluating AI research agents.

What you can do with it

  • Train agents on diverse research tasks
  • Evaluate reinforcement learning algorithms
  • Analyse agent behaviour with trajectory visualisation
  • Compare how agents generalise across domains

Getting started

  • Clone the repository and install dependencies
  • Create an env file with keys and paths
  • Install Docker or Podman and pull an image
  • Launch a task with the run.py script

Pros & cons

✓ What we like

  • Docker and Podman containers support reproducible runs
  • Installation and usage guides are provided

! What to watch out for

  • Setup involves cloning a repo and preparing an env file
  • Runs need a container runtime such as Docker or Podman

FAQ

Who developed it?

The material names the Meta GenAI team and the UCSB NLP team.

How many tasks are included?

The material states 13 diverse AI research tasks.

Last reviewed: 2026-11-15

More LLM API platform tools

View all →

How we review