AgentGym-RL and AgentGym
These are successive versions of the same research project, with AgentGym-RL focusing specifically on multi-turn reinforcement learning for long-horizon tasks while the newer AgentGym broadens the scope to agent training across diverse environments.
About AgentGym-RL
WooooDyy/AgentGym-RL
Code and implementations for the paper "AgentGym-RL: Training LLM Agents for Long-Horizon Decision Making through Multi-Turn Reinforcement Learning" by Zhiheng Xi et al.
About AgentGym
WooooDyy/AgentGym
Code and implementations for the ACL 2025 paper "AgentGym: Evolving Large Language Model-based Agents across Diverse Environments" by Zhiheng Xi et al.
Integrates 14 diverse environments—web navigation, text games, embodied tasks, and SQL reasoning—standardized via ReAct format with HTTP-based environment servers enabling modular deployment and custom environment development. The AgentEvol method trains agents through trajectory-based self-evolution, supported by AgentTraj-L trajectory dataset and AgentEval benchmark; includes RL framework variant (AgentGym-RL) for direct interactive learning with multi-turn reinforcement learning on long-horizon tasks.
Scores updated daily from GitHub, PyPI, and npm data. How scores work