The Year in AI Papers
The year/Independent research

Paper 2604.18292

Agent-World: Scaling Real-World Environment Synthesis for Evolving General Agent Intelligence

Guanting Dong, Junting Lu, Junjie Huang, Wanjun Zhong, Longxiang Liu, Shijue Huang, Zhenyu Li, Yang Zhao, Xiaoshuai Song, Xiaoxi Li, Jiajie Jin, Yutao Zhu, et al.

Published
Apr 2026
Research lab
Independent
Citations
14
GitHub
Not linked

01 In brief

Summary

Agent-World is a self-evolving training arena for general-purpose AI agents, developed by Renmin University of China and ByteDance Seed.

It addresses two key bottlenecks in agent training: scalable realism in environment synthesis and continuous self-improvement.

The system has two main components: (1) Agentic Environment-Task Discovery, which autonomously mines topic-aligned databases and executable tools from real-world sources (MCP servers, tool docs, PRDs), creating 1,978 environments and 19,822 tools, and synthesizes verifiable tasks with controllable difficulty via graph-based and programmatic methods; (2) Continuous Self-Evolving Agent Training, which uses multi-environment reinforcement learning (GRPO) with executable rewards, and a self-evolving arena that diagnoses capability gaps and drives targeted data expansion.

Evaluated on 23 benchmarks, Agent-World-8B and 14B consistently outperform strong baselines, including proprietary models like GPT-5.2 and Gemini-3 Pro on several tasks.

Scaling analysis shows performance improves with environment diversity and self-evolution rounds, with diminishing returns at larger scales.

The self-evolving loop yields monotonic gains, particularly on MCP-Mark, and benefits other baselines like EnvScaler-8B.

02 From the paper

Abstract

Large language models are increasingly expected to serve as general-purpose agents that interact with external, stateful tool environments. The Model Context Protocol (MCP) and broader agent skills offer a unified interface for connecting agents with scalable real-world services, but training robust agents remains limited by the lack of realistic environments and principled mechanisms for life-long learning. In this paper, we present \textbf{Agent-World}, a self-evolving training arena for advancing general agent intelligence through scalable environments. Agent-World has two main components: (1) Agentic Environment-Task Discovery, which autonomously explores topic-aligned databases and executable tool ecosystems from thousands of real-world environment themes and synthesizes verifiable tasks with controllable difficulty; and (2) Continuous Self-Evolving Agent Training, which combines multi-environment reinforcement learning with a self-evolving agent arena that automatically identifies capability gaps through dynamic task synthesis and drives targeted learning, enabling the co-evolution of agent policies and environments. Across 23 challenging agent benchmarks, Agent-World-8B and 14B consistently outperforms strong proprietary models and environment scaling baselines. Further analyses reveal scaling trends in relation to environment diversity and self-evolution rounds, offering insights for building general agent intelligence.