AgentGym-RL: Training Data for Long-Horizon LLM Agent Decision Making

AgentGym-RL is a dataset and benchmark for training Large Language Model agents through multi-turn reinforcement learning. It was created by the AgentGym organization and last updated on September 12, 2025. The data is designed to address long-horizon decision-making tasks.

Use Cases

• Training reinforcement learning models for long-horizon planning based on multi-turn interactions.
• Benchmarking LLM agent performance on sequential decision-making tasks.
• Developing and evaluating new RL algorithms for language model agents.

Strengths

• Dataset is directly associated with a published research paper (arXiv:2509.08755).
• Last updated on 2025-09-12, suggesting recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

AgentGym organization

Collection Method

Likely generated through multi-turn reinforcement learning experiments as described in the associated paper.

Freshness

Last updated 2025-09-12 15:09:45