SWE-Universe: Scale Real-World Verifiable Environments to Millions
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , , , , , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2026
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866912868785455104 |
|---|---|
| author | Chen, Mouxiang Zhang, Lei Feng, Yunlong Wang, Xuwu Zhao, Wenting Cao, Ruisheng Yang, Jiaxi Chen, Jiawei Li, Mingze Ma, Zeyao Ge, Hao Zhang, Zongmeng Cui, Zeyu Liu, Dayiheng Zhou, Jingren Sun, Jianling Lin, Junyang Hui, Binyuan |
| author_facet | Chen, Mouxiang Zhang, Lei Feng, Yunlong Wang, Xuwu Zhao, Wenting Cao, Ruisheng Yang, Jiaxi Chen, Jiawei Li, Mingze Ma, Zeyao Ge, Hao Zhang, Zongmeng Cui, Zeyu Liu, Dayiheng Zhou, Jingren Sun, Jianling Lin, Junyang Hui, Binyuan |
| contents | We propose SWE-Universe, a scalable and efficient framework for automatically constructing real-world software engineering (SWE) verifiable environments from GitHub pull requests (PRs). To overcome the prevalent challenges of automatic building, such as low production yield, weak verifiers, and prohibitive cost, our framework utilizes a building agent powered by an efficient custom-trained model. This agent employs iterative self-verification and in-loop hacking detection to ensure the reliable generation of high-fidelity, verifiable tasks. Using this method, we scale the number of real-world multilingual SWE environments to a million scale (807,693). We demonstrate the profound value of our environments through large-scale agentic mid-training and reinforcement learning. Finally, we applied this technique to Qwen3-Max-Thinking and achieved a score of 75.3% on SWE-Bench Verified. Our work provides both a critical resource and a robust methodology to advance the next generation of coding agents. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_02361 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | SWE-Universe: Scale Real-World Verifiable Environments to Millions Chen, Mouxiang Zhang, Lei Feng, Yunlong Wang, Xuwu Zhao, Wenting Cao, Ruisheng Yang, Jiaxi Chen, Jiawei Li, Mingze Ma, Zeyao Ge, Hao Zhang, Zongmeng Cui, Zeyu Liu, Dayiheng Zhou, Jingren Sun, Jianling Lin, Junyang Hui, Binyuan Software Engineering Artificial Intelligence We propose SWE-Universe, a scalable and efficient framework for automatically constructing real-world software engineering (SWE) verifiable environments from GitHub pull requests (PRs). To overcome the prevalent challenges of automatic building, such as low production yield, weak verifiers, and prohibitive cost, our framework utilizes a building agent powered by an efficient custom-trained model. This agent employs iterative self-verification and in-loop hacking detection to ensure the reliable generation of high-fidelity, verifiable tasks. Using this method, we scale the number of real-world multilingual SWE environments to a million scale (807,693). We demonstrate the profound value of our environments through large-scale agentic mid-training and reinforcement learning. Finally, we applied this technique to Qwen3-Max-Thinking and achieved a score of 75.3% on SWE-Bench Verified. Our work provides both a critical resource and a robust methodology to advance the next generation of coding agents. |
| title | SWE-Universe: Scale Real-World Verifiable Environments to Millions |
| topic | Software Engineering Artificial Intelligence |
| url | https://arxiv.org/abs/2602.02361 |