2602.08990v1 Feb 09, 2026 cs.AI

InternAgent-1.5: 장기 자율 과학 발견을 위한 통합 에이전트 프레임워크

InternAgent-1.5: A Unified Agentic Framework for Long-Horizon Autonomous Scientific Discovery

Shuyue Hu
Shuyue Hu
Citations: 91
h-index: 5
Peng Ye
Peng Ye
Citations: 42
h-index: 4
Wenjie Lou
Wenjie Lou
Citations: 10
h-index: 1
Lilong Wang
Lilong Wang
Citations: 35
h-index: 4
Tianshuo Peng
Tianshuo Peng
Citations: 381
h-index: 7
Songtao Huang
Songtao Huang
Citations: 125
h-index: 3
Dongrui Liu
Dongrui Liu
Citations: 2
h-index: 1
Jiakang Yuan
Jiakang Yuan
Fudan University
Citations: 432
h-index: 12
Shiyang Feng
Shiyang Feng
Citations: 88
h-index: 5
Runmin Ma
Runmin Ma
Citations: 84
h-index: 4
Xiang-yu Yan
Xiang-yu Yan
Citations: 9
h-index: 2
Yue Fan
Yue Fan
Citations: 11
h-index: 2
Yusong Hu
Yusong Hu
Citations: 30
h-index: 2
Zongsheng Cao
Zongsheng Cao
Citations: 8
h-index: 1
Zi-Wen Guo
Zi-Wen Guo
Citations: 8
h-index: 1
Shangheng Du
Shangheng Du
Citations: 27
h-index: 2
Weida Wang
Weida Wang
Shanghai AI Laboratory
Citations: 96
h-index: 5
Jinxin Shi
Jinxin Shi
Citations: 6
h-index: 1
Yuhao Zhou
Yuhao Zhou
Citations: 50
h-index: 5
Xiaohan He
Xiaohan He
Citations: 28
h-index: 2
Zhiyin Yu
Zhiyin Yu
Citations: 31
h-index: 2
Fangchen Yu
Fangchen Yu
Citations: 25
h-index: 3
Qihao Zheng
Qihao Zheng
Citations: 80
h-index: 6
Jiamin Wu
Jiamin Wu
Citations: 180
h-index: 4
Mianxin Liu
Mianxin Liu
Citations: 81
h-index: 6
Chi Zhang
Chi Zhang
Citations: 1
h-index: 1
Shaowei Hou
Shaowei Hou
Citations: 53
h-index: 2
Shuya Li
Shuya Li
Citations: 40
h-index: 4
Y. Jiang
Y. Jiang
Citations: 1
h-index: 1
Zifu Wang
Zifu Wang
Citations: 8
h-index: 1
Jiong-hui Wang
Jiong-hui Wang
Citations: 4
h-index: 1
Wang-Sheng Xu
Wang-Sheng Xu
Citations: 2
h-index: 1
Yue Deng
Yue Deng
Citations: 83
h-index: 5
Yiheng Wang
Yiheng Wang
Citations: 144
h-index: 5
Wenlong Zhang
Wenlong Zhang
Citations: 182
h-index: 8
Fenghua Ling
Fenghua Ling
Citations: 773
h-index: 14
Shufei Zhang
Shufei Zhang
Citations: 28
h-index: 2
Xiaosong Wang
Xiaosong Wang
Citations: 29
h-index: 1
Shu-qi Zheng
Shu-qi Zheng
Citations: 361
h-index: 3
Siqi Sun
Siqi Sun
Citations: 44
h-index: 2
Chun-dong Song
Chun-dong Song
Citations: 3
h-index: 1
Bin Wang
Bin Wang
Citations: 68
h-index: 3
Conghui He
Conghui He
Citations: 23
h-index: 3
Yihao Liu
Yihao Liu
Citations: 1
h-index: 1
Xin Li
Xin Li
Citations: 5
h-index: 1
Q. Hou
Q. Hou
Citations: 71
h-index: 3
Tao Chen
Tao Chen
Citations: 2
h-index: 1
Xiangyu Yue
Xiangyu Yue
Citations: 18
h-index: 3
Liang He
Liang He
Citations: 29
h-index: 3
Dahua Lin
Dahua Lin
Citations: 1,380
h-index: 8
Bowen Zhou
Bowen Zhou
Citations: 104
h-index: 5
Lei Bai
Lei Bai
Citations: 73
h-index: 5
Shuai Zhang
Shuai Zhang
Citations: 38
h-index: 4
Zhijie Zhong
Zhijie Zhong
Citations: 156
h-index: 4
Xun Huang
Xun Huang
Citations: 51
h-index: 2
Bo Zhang
Bo Zhang
Citations: 2
h-index: 1

우리는 계산 및 실증 영역 전반에 걸친 엔드투엔드 과학 발견을 위해 설계된 통합 시스템인 InternAgent-1.5를 소개합니다. 이 시스템은 생성, 검증, 진화라는 세 가지 조정된 하위 시스템으로 구성된 구조화된 아키텍처를 기반으로 합니다. 이러한 하위 시스템들은 심층 연구, 솔루션 최적화, 장기 기억(long horizon memory)을 위한 기반 능력에 의해 지원됩니다. 이 아키텍처를 통해 InternAgent-1.5는 일관되고 발전하는 행동을 유지하면서 확장된 발견 주기 동안 지속적으로 작동할 수 있습니다. 또한 단일 통합 시스템 내에서 계산 모델링과 실험실 실험을 조정할 수 있게 합니다. 우리는 GAIA, HLE, GPQA, FrontierScience와 같은 과학적 추론 벤치마크에서 InternAgent-1.5를 평가했으며, 이 시스템은 강력한 기반 능력을 입증하는 선도적인 성능을 달성했습니다. 이러한 벤치마크를 넘어, 우리는 두 가지 범주의 발견 과제를 추가로 평가했습니다. 알고리즘 발견 과제에서 InternAgent-1.5는 핵심 머신 러닝 문제를 위한 경쟁력 있는 방법론을 자율적으로 설계합니다. 실증적 발견 과제에서는 완전한 계산 실험 또는 습식 실험(wet lab)을 수행하고 지구, 생명, 생물 및 물리 영역에서 과학적 발견을 도출합니다. 전반적으로 이러한 결과는 InternAgent-1.5가 자율 과학 발견을 위한 일반적이고 확장 가능한 프레임워크를 제공함을 보여줍니다.

Original Abstract

We introduce InternAgent-1.5, a unified system designed for end-to-end scientific discovery across computational and empirical domains. The system is built on a structured architecture composed of three coordinated subsystems for generation, verification, and evolution. These subsystems are supported by foundational capabilities for deep research, solution optimization, and long horizon memory. The architecture allows InternAgent-1.5 to operate continuously across extended discovery cycles while maintaining coherent and improving behavior. It also enables the system to coordinate computational modeling and laboratory experimentation within a single unified system. We evaluate InternAgent-1.5 on scientific reasoning benchmarks such as GAIA, HLE, GPQA, and FrontierScience, and the system achieves leading performance that demonstrates strong foundational capabilities. Beyond these benchmarks, we further assess two categories of discovery tasks. In algorithm discovery tasks, InternAgent-1.5 autonomously designs competitive methods for core machine learning problems. In empirical discovery tasks, it executes complete computational or wet lab experiments and produces scientific findings in earth, life, biological, and physical domains. Overall, these results show that InternAgent-1.5 provides a general and scalable framework for autonomous scientific discovery.

1 Citations
0 Influential
7 Altmetric
36.0 Score

No Analysis Report Yet

This paper hasn't been analyzed by Gemini yet.

댓글

댓글을 작성하려면 로그인하세요.

아직 댓글이 없습니다. 첫 번째 댓글을 남겨보세요!