Immersion in the GitHub Universe: Scaling Coding Agents to Mastery ∗ Equal Contributions. † Corresponding authors. Date: Feb. 10, 2026.
Jiale Zhao, Guoxin Chen, Fanzhe Meng, Minghao Li, Jie Chen, Hui Xu, Yongshuai Sun, Wayne Xin Zhao, Ruihua Song, Yuan Zhang, Peng Wang, Cheng Chen, Ji-Rong Wen, Kai Jia · 2026
Read Immersion in the GitHub Universe: Scaling Coding Agents to Mastery ∗ Equal Contributions. † Corresponding authors. Date: Feb. 10, 2026. by Jiale Zhao, Guoxin Chen, Fanzhe Meng, Minghao Li, Jie Chen, Hui Xu, Yongshuai Sun, Wayne Xin Zhao, Ruihua Song, Yuan Zhang, Peng Wang, Cheng Chen, Ji-Rong Wen, Kai Jia (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-09892.txt.
Description
High-quality training data is the bottleneck for coding agents that aim at real-world software mastery, since environment setup, test generation, and problem curation are all hard to scale. Jiale Zhao and thirteen co-authors propose ScaleSWE, an automated, sandboxed multi-agent workflow for building such data at scale, in their 2026 paper. Three specialized agents handle environment setup, test creation, and problem description synthesis.
Running over six million pull requests across fifty-two hundred repositories, the system produced ScaleSWE Data. It holds one hundred thousand verified instances, the largest such dataset to date, with broad repository diversity and realistic task complexity. The team distilled seventy-one thousand four hundred ninety-eight high-quality trajectories and fine-tuned a model to produce ScaleSWE Agent. It achieves a sixty-four percent resolve rate on SWE-Bench Verified, nearly triple the base model. ScaleSWE will be publicly available.
Catalog record
| CREATOR | Jiale Zhao, Guoxin Chen, Fanzhe Meng, Minghao Li, Jie Chen, Hui Xu, Yongshuai Sun, Wayne Xin Zhao, Ruihua Song, Yuan Zhang, Peng Wang, Cheng Chen, Ji-Rong Wen, Kai Jia |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 61 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-09892