From SWE-Zero to SWE-Hero : Execution-free to Execution-based Fine-tuning for Software Engineering Agents

Nikolai Ludwig, Wasi Uddin Ahmad, Somshubra Majumdar, Boris Ginsburg · 2026

Read From SWE-Zero to SWE-Hero : Execution-free to Execution-based Fine-tuning for Software Engineering Agents by Nikolai Ludwig, Wasi Uddin Ahmad, Somshubra Majumdar, Boris Ginsburg (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-2604-01496.txt.

Description

Ludwig and colleagues introduce SWE-ZERO to SWE-HERO, a two-stage fine-tuning recipe for SWE-bench distilling frontier language models. SWE-ZERO uses execution-free trajectories to master code semantics and repository reasoning. SWE-HERO applies execution-backed refinement to turn semantic intuitions into rigorous engineering workflows. They release 300k SWE-ZERO and 13k SWE-HERO trajectories from Qwen3-Coder-480B, plus agents. SWE-HERO-32B achieves a 62.2 percent resolution rate on SWE-bench Verified. Despite training only on Python, the agents transfer to SWE-bench Multilingual.

Catalog record

CREATORNikolai Ludwig, Wasi Uddin Ahmad, Somshubra Majumdar, Boris Ginsburg
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE48394
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.01496

Browse the full library · About the Tower of Babel