AI Paper Atlas

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

ICLR 2026 · Reinforcement learning › Visual and representation RL · approach: In-context and adaptive RL

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 changedWhereMeasuredEffect
add ScalingInter-RL progressive horizon scaling to PPO (ablation)AgentGym-RL-7B (PPO) · SciWorldscore: 10.83 -> 25.69Improved
add ScalingInter-RL to PPO (ablation)AgentGym-RL-7B (PPO) · TextCraftsuccess rate: 68.00 -> 71.00Improved
add ScalingInter-RL to PPO (ablation)AgentGym-RL-7B (PPO) · BabyAIsuccess rate: 86.66 -> 90.00Improved
add ScalingInter-RL to REINFORCE++ (ablation)AgentGym-RL-7B (REINFORCE++) · SciWorldscore: 13.63 -> 26.14Improved
add ScalingInter-RL to REINFORCE++ (ablation)AgentGym-RL-7B (REINFORCE++) · TextCraftsuccess rate: 73.00 -> 77.00Improved
use GRPO instead of REINFORCE++ (ablation)Qwen2.5-7B-Instruct agent · TextCraftsuccess rate: 73.00 -> 89.00Improved
use GRPO instead of REINFORCE++ (ablation)Qwen2.5-3B-Instruct agent · TextCraftsuccess rate: 28.00 -> 75.00Improved
use GRPO instead of REINFORCE++ (ablation)Qwen2.5-7B-Instruct agent · SearchQAaccuracy: 24.00 -> 34.00Improved
use GRPO instead of REINFORCE++ (ablation)Qwen2.5-3B-Instruct agent · BabyAIsuccess rate: 70.00 -> 93.33Improved
vary stage transition frequency from 100 to 75 or 125 (ablation)ScalingInter-RL turns [5,8,10] · Deep Searchperformance: 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 Searchperformance: 38.3 -> 36.8No clear effect
change interaction turn list from [5,8,10] to [5,10,15] (ablation)ScalingInter-RL · Deep Searchperformance: 38.3 -> 39.1No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy