AI Paper Atlas

Research map / Multi-agent systems and games / LLM agents

GEM: A Gym for Generalist LLMs

Zichen Liu, Anya Sims, Keyu Duan, Changyu Chen, Simon Yu, Xiangxin Zhou, Haotian Xu, Shaopan Xiong, Bo Liu, Chenmien Tan, Weixun Wang, Hao Zhu and 6 more

ICLR 2026 · Multi-agent systems and games › LLM agents · approach: Agent data synthesis and behavior studies

Read the paperCheck your idea against 137,217 papers

AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Introduces GEM, an open-source Gym-style environment simulator for training and evaluating LLM agents via experience-based learning.

Problem

LLM training is shifting from static datasets to experience-based learning in interactive environments, but there is no standard environment framework analogous to OpenAI Gym for LLM agents.

Approach

An open-source environment simulator designed for LLMs, analogous to OpenAI-Gym for traditional RL, providing a standardized interface for agent environments to generate experience for training generalist LLMs.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add Return Batch Normalization to REINFORCE (ablation)REINFORCE (Qwen3 agents) · 24 GEM environmentsepisode returnImproved
use GRPO instead of REINFORCE/PPO in multi-turn environments with dense rewardsGRPO · GuessTheNumber, Sudokuepisode returnHurt
use GRPO in single-turn ReasoningGym tasksGRPO · ReasoningGym tasksepisode returnNo clear effect
use critic-based PPO instead of ReBNPPO · Sudoku vs Minesweeperepisode returnMixed
lower discount factor gamma (0.999 to 0.9) (ablation)REINFORCE+ReBN · GuessTheNumberturns to solve: converges toward optimal about 5.6 turnsImproved
add Python tool to RL training (ablation)Qwen3-4B-Base · math average (5 sets)Pass@1: 41.4 -> 49.8Improved
RL training without tool (ablation)Qwen3-4B-Base · math averagePass@1: 35.3 -> 41.4Improved
give base model Python tool without RL (ablation)Qwen3-4B-Base · MinervaMathPass@1: 36.4 -> 30.1Hurt
add Search tool to RL training (ablation)Qwen3-4B · QA average (7 sets)QA score: 23.9 -> 45.5Improved
train on mixed HotpotQA+NQ instead of HotpotQA only (ablation)Qwen3-4B with search · QA averageQA score: 43.7 -> 45.5Improved
use fixed negative reward (-1) instead of ReBN (ablation)REINFORCE · math:Orz57K / qa:HotpotQAtool useHurt
use vanilla REINFORCE with 0/1 reward (no ReBN) (ablation)REINFORCE · multi-turn tool-usetool usageHurt

Experimental setup

Builds on: OpenAI Gym

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