Aha.
正在载入中英对照阅读…

arXiv:2604.17406 · 中英对照阅读

EvoMaster:面向规模化智能体科学的基础性演化智能体框架

EvoMaster: A Foundational Evolving Agent Framework for Agentic Science at Scale

Xinyu Zhu、Yuzhu Cai、Zexi Liu、Cheng Wang、Fengyang Li、Wenkai Jin、Wanxu Liu、Zehao Bing、Bingyang Zheng、Jingyi Chai、Shuo Tang、Rui Ye、Yuwen Du、Xianghe Pang、Yaxin Du、Tingjia Miao、Yuzhi Zhang、Ruoxue Liao、Zhaohan Ding、Linfeng Zhang、Yanfeng Wang、Weinan E、Siheng Chen

中文速览

科研智能体常常在不同领域重复搭建工具,而且一项研究里的证据和经验很难留给下一项研究继续用。EvoMaster 用三层相扣的循环来解决这两个问题:在单次任务中边行动边观察,在一项研究中根据实验结果调整假设,再把多项研究中积累的经验整理成可复用的能力。它在十项科学研究、编程和推理基准中有九项得分领先,平均得分为 58.02%,高于最强对手 Codex 的 40.29%,成本还低 35.6%。这说明,共用一套能持续积累证据和经验的基础架构,有望让智能体更广泛地参与科研,也让研究成果更容易推动后续研究。

摘要

大型语言模型与智能体的融合正在催生科学发现的新时代:智能体科学。然而,不同科学领域反复重建通用智能体基础设施(循环碎片化),而长周期研究中的有用证据与经验则会丢失(循环不连续),为规模化智能体科学带来障碍。我们提出 EvoMaster,一种面向规模化智能体科学的基础性演化智能体框架。EvoMaster 通过实现循环研究来应对循环碎片化和循环不连续,在循环研究中,外部证据得以保留,并用于改进后续决策。通过执行、探索和演化三个嵌套循环(E3(三重循环)),循环研究将单次运行内、跨实验以及跨研究的研究工作连接起来。在涵盖科学研究、编程和推理的十项基准测试中,EvoMaster 在使用 GPT-5.4 的四个智能体中取得最高分,平均得分达到 58.02%,并以低 35.6%的成本超越了最强竞争智能体 Codex(40.29%)。这些结果表明,共享的循环研究基础能够支持多样化的科学智能体实现规模化发展。

The convergence of large language models and agents is catalyzing a new era of scientific discovery: Agentic Science. However, common agent infrastructure is repeatedly rebuilt across scientific fields (loop fragmentation) and useful evidence and experience are lost in long-horizon research (loop discontinuity), bringing obstacles to Agentic Science at Scale. We introduce EvoMaster, a foundational evolving agent framework for Agentic Science at Scale. EvoMaster handles loop fragmentation and loop discontinuity by implementing Loop Research in which external evidence persists and improves later decisions. Through three nested loops, Execution, Exploration and Evolution (E3), loop research connects research within runs, across experiments and across studies. Across ten benchmarks spanning scientific research, coding and reasoning, EvoMaster achieves the best score among four agents using GPT-5.4, reaching a mean score of 58.02%, and outperforms the strongest competing agent Codex(40.29%) while costing 35.6% less. These results show that a shared loop-research foundation can support diverse scientific agents at scale.

术语表

Agentic Science
智能体科学
Agentic Science at Scale
规模化智能体科学
Auto Research
自动化研究
Loop Research
循环研究
EvoMaster
EvoMaster
SciMaster
SciMaster
E3
E3(三重循环)
Execution Loop
执行循环
Exploration Loop
探索循环
Evolution Loop
演化循环
loop fragmentation
循环碎片化
loop discontinuity
循环不连续
foundational evolving agent framework
基础性演化智能体框架
Playground–Exp–Agent architecture
Playground–Exp–Agent 架构
persistent research state
持久化研究状态
evidence-sensitive persistent state
证据敏感型持久状态
structured trajectories
结构化轨迹
evaluator-guided updates
评估器引导的更新
experimental run
实验运行
research campaign
研究周期
Execution-level autonomy
执行级自主性
Exploration-level autonomy
探索级自主性
Evolution-level autonomy
演化级自主性
foundation-model agents
基础模型智能体
tool orchestration
工具编排
context management
上下文管理
backend substitution
后端替换
GPT-5.4
GPT-5.4
Codex
Codex
AlphaEvolve
AlphaEvolve