# openai/procgen

Procgen Benchmark: Procedurally-Generated Game-Like Gym-Environments

Repository: https://github.com/openai/procgen
Canonical: https://ross.abutalabs.com/products/procgen
Homepage: https://openai.com/blog/procgen-benchmark/
Language: C++
License: MIT
License Family: permissive
Last push: 2026-03-27T02:52:42+00:00

## Health v2 (maintenance only)
Score: 56/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 74, release rhythm 8, longevity 100
- inputs: {"age_days": 2477, "days_push": 159, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1185, forks 222 (observed 2026-08-28T04:03:54.872180+00:00)

## What it is
A benchmark of 16 procedurally-generated, game-like Gym environments for measuring how quickly reinforcement learning agents learn generalizable skills. Environments run at thousands of steps per second on a single core and are customizable when built from source.

## Use cases
- benchmark reinforcement learning generalization
- measure RL sample efficiency
- train RL agents on procedurally generated levels
- run fast gym environments for RL research
- compare RL agents against overfitting to fixed levels
- build custom RL environments from source

## When to choose
- You need fast, randomized Gym environments to test RL generalization
- You want a standard benchmark comparable to published Procgen results
- You need environments that are easy to modify for RL experiments

## When to avoid
- You need actively developed features or new environments
- You require GPU-accelerated or vectorized-only environments
- You need environments beyond Python 3.10 or without AVX CPU support

## Facets
- artifact type: library
- maturity: maintenance
- function: simulation, machine-learning, reinforcement-learning, game-engine
- domain: reinforcement-learning, machine-learning, gaming-tools
- platform: python, windows, cpp
- tags: gym-environments, procedural-generation, rl-benchmark, openai, research, linux, macos

## Member repositories
- openai/procgen (main) score 56

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:54.872180+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:24:08.309111+00:00, confidence not recorded.
  - readme: https://github.com/openai/procgen (fetched 2026-08-28T04:03:54.872180+00:00, sha 63437dfe9cb5)
- Data as of 2026-08-30T08:39:29.467469+00:00.