OPEN SOURCE & RESEARCH

Build on what the community has made.

A carefully selected starting set of public frameworks and benchmarks. Always review original documentation, current licenses, and implementation details before adopting a project.

VE↗

LLM training and evaluation environments

Verifiers

An open-source library for building environments for evaluating and training language models.

Visit official resource ↗

NV↗

Agent evaluation and training infrastructure

NVIDIA NeMo Gym

Environment infrastructure with task datasets, agent harnesses, verification, and state.

Visit official resource ↗

BR↗

Browser agent benchmarks

BrowserGym

Extensible framework spanning browser tasks and benchmark integrations.

Visit official resource ↗

SW↗

Software engineering evaluation

SWE-bench

Benchmark family for software engineering agents working against real repository issues.

Visit official resource ↗

OS↗

Desktop computer-use benchmark

OSWorld

Benchmark for agents working in real desktop computer environments.

Visit official resource ↗

GY↗

General RL environment interface

Gymnasium

Common API for single-agent reinforcement learning environments.

Visit official resource ↗

PE↗

Multi-agent RL interface

PettingZoo

Environment interfaces for multi-agent reinforcement learning research.

Visit official resource ↗

Τ-↗

Tool-using customer support agents

τ-bench

Research benchmark around tool-using conversational agents and policy-constrained tasks.

Visit official resource ↗

Editorial note. These are external projects, not products sold or operated by RLenvironment.ai. Inclusion is informational, not a partnership or endorsement.