ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments
Turn a real scientific codebase into a row of workbenches that run, get scored and can be repeated, so an AI can work on real scientific tasks and be checked.
Move the cursor over the diagram
Overview
Science integrated development platform
Infrastructure for scientific experience: real scientific code, tasks, environments and reinforcement-learning training.
ScienceIDE is an independent infrastructure exploration for real scientific codebases, research tasks and verifiable scientific experience.
It turns real scientific code, research procedures, task goals and verification criteria into scientific environments that run, measure and reproduce, so that researchers can do agent execution, model training and reinforcement-learning research in real scientific settings.
Its value is in organising real scientific software, domain judgement, task goals and verifiable feedback into one executable environment, an experiential base for research agents, model training, reinforcement learning and discovery research. ScienceIDE does not require ScienceBuddy to be used, and does not depend on JEPA-Anything to stand.
Current task forms
- Accelerate
- Discover
- Repair
- Reproduce
- Integrate
The project page, the technical report and the PhAI-IDE model weights are public. The models are trained on interaction trajectories verified in ScienceIDE, which remains open to task proposals.
Abstract
Scientific code repositories encode decades of human knowledge in executable models, methods, and tools. Yet fragmented toolchains, implicit domain conventions, and specialized correctness criteria make this knowledge difficult to convert into reliable learning experience—a challenge we call the scientific experience bottleneck. We introduce ScienceIDE, infrastructure for turning the world’s scientific code into programmable environments for scientific agents. Guided by expert-defined scientific cases and acceptance criteria, agents transform repositories into executable environments that support task generation, execution, and scientific verification. These environments provide a shared foundation for supervised fine-tuning, reinforcement learning, and evaluation. Using verified interaction trajectories, we train PhAI-IDE-72B, PhAI-IDE-9B, and PhAI-IDE-4B. The model family shows gains in held-out scientific-code repair and across selected general-purpose benchmarks in code, reasoning, and knowledge, providing evidence of positive transfer from scientific experience to broader capabilities. ScienceIDE lays the foundation for an integrated workspace for agent learning and scientific practice, making humanity’s scientific software a shared substrate for developing scientific intelligence.
Authors
* Equal contribution
Venue
Cite
BibTeX
@article{geng2026scienceide,
title={ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments},
author={Geng, Hejia and Huang, Zesen and Li, Haoyang and Li, Wenbin and Wu, Koutian and Zhou, Zihan and Pang, Yuanbo and Liu, Weihao and Xu, Zigong and Li, Zhiping and Zhang, Zongzheng and Dong, Chuanfei and Sun, Jiankai and Zheng, Tianzhe and Xie, Fengyu and Ma, Yue and Shi, Yueheng and Wang, Junkai and Xie, Tong and Di, Zonglin and Liu, Xianrong and Gao, Qucheng and Liu, Yimin and Pan, Jiaming and Huang, Sheng and Ma, Xiao-Han and Yuan, Lanqing and Zhu, Zhenlin and Liu, Ziang and Xu, Ziyang and Liang, Kangkai and Xian, Jiayi and Zhao, Zehong and Ma, Taigao and Chang, Fuming and Hu, Yuanzhe and Xu, Liuwei and Tang, Xianzhe and Xie, Jingxu and Zhang, Peijin and Gao, Qiang and Xing, Chengyi and Zhao, Zhe and Wang, Xi and Wang, Leo and Wu, Fang and Xing, Yaopeng and Meng, Xing and Yin, Zhenfei and Wu, Yingcheng and Yang, Ling},
journal={arXiv preprint arXiv:2609.19134},
year={2026}
}