AgentGym

AgentGym is a benchmark containing eight environments used to evaluate the performance of agents, specifically in terms of task completion rates and efficiency in decision-making.

Papers using AgentGym (4)

Agentic Monte Carlo: Simulating Reinforcement Learning for Black-Box Agents2026PatchWorld: Gradient-Free Optimization of Executable World Models for Agent Environments2026PABU: Progress-aware Belief Update For Efficient LLM Agents2026R2e-gym: Procedural Environments And Hybrid Verifiers For Scaling Open-weights SWE Agents2025