Research map / Reinforcement learning / Visual and representation RL
AgentGym-RL: An Open-Source Framework to Train LLM Agents for Long-Horizon Decision Making via Multi-Turn RL
Zhiheng Xi, Jixuan Huang, Chenyang Liao, Baodai Huang, Jiaqi Liu, Honglin Guo, yajie yang, Rui Zheng, Junjie Ye, Jiazheng Zhang, Wenxiang Chen, Wei He and 12 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add ScalingInter-RL progressive horizon scaling to PPO (ablation) | AgentGym-RL-7B (PPO) · SciWorld | score: 10.83 -> 25.69 | Improved |
| add ScalingInter-RL to PPO (ablation) | AgentGym-RL-7B (PPO) · TextCraft | success rate: 68.00 -> 71.00 | Improved |
| add ScalingInter-RL to PPO (ablation) | AgentGym-RL-7B (PPO) · BabyAI | success rate: 86.66 -> 90.00 | Improved |
| add ScalingInter-RL to REINFORCE++ (ablation) | AgentGym-RL-7B (REINFORCE++) · SciWorld | score: 13.63 -> 26.14 | Improved |
| add ScalingInter-RL to REINFORCE++ (ablation) | AgentGym-RL-7B (REINFORCE++) · TextCraft | success rate: 73.00 -> 77.00 | Improved |
| use GRPO instead of REINFORCE++ (ablation) | Qwen2.5-7B-Instruct agent · TextCraft | success rate: 73.00 -> 89.00 | Improved |
| use GRPO instead of REINFORCE++ (ablation) | Qwen2.5-3B-Instruct agent · TextCraft | success rate: 28.00 -> 75.00 | Improved |
| use GRPO instead of REINFORCE++ (ablation) | Qwen2.5-7B-Instruct agent · SearchQA | accuracy: 24.00 -> 34.00 | Improved |
| use GRPO instead of REINFORCE++ (ablation) | Qwen2.5-3B-Instruct agent · BabyAI | success rate: 70.00 -> 93.33 | Improved |
| vary stage transition frequency from 100 to 75 or 125 (ablation) | ScalingInter-RL turns [5,8,10] · Deep Search | performance: 38.3 -> 37.8 (75), 38.5 (125) | No clear effect |
| change interaction turn list from [5,8,10] to [3,8,13] (ablation) | ScalingInter-RL · Deep Search | performance: 38.3 -> 36.8 | No clear effect |
| change interaction turn list from [5,8,10] to [5,10,15] (ablation) | ScalingInter-RL · Deep Search | performance: 38.3 -> 39.1 | No clear effect |
Closest papers by meaning
- Dr. MAS: Stable Reinforcement Learning for Multi-Agent LLM Systems (NeurIPS 2026 · similarity 0.91)
- Generalizable End-to-End Tool-Use RL with Synthetic CodeGym (ICLR 2026 · similarity 0.90)
- DRAFT-RL: Multi-Agent Chain-of-Draft Reasoning for Reinforcement Learning-Enhanced LLMs (AAAI 2026 · similarity 0.90)
- AstraFlow: Dataflow-Oriented Reinforcement Learning for Agentic LLMs (NeurIPS 2026 · similarity 0.90)
- Dynamic Optimizations of LLM Ensembles with Two-Stage Reinforcement Learning Agents (ICML 2026 · similarity 0.90)
- Skill-SD: Skill-Conditioned Self-Distillation for Multi-turn LLM Agents (NeurIPS 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
