Researchers have developed OpenSWE, a large-scale, transparent framework for training software engineering (SWE) agents. This framework includes over 45,000 executable Docker environments derived from more than 12,000 repositories, with all associated code and infrastructure publicly available for reproducibility. The project involved a multi-agent synthesis pipeline and a quality-filtering process to ensure environment effectiveness, costing approximately $1.47 million in total. Experiments show that models trained with OpenSWE, such as OpenSWE-32B and OpenSWE-72B, achieve state-of-the-art results on the SWE-bench Verified benchmark and demonstrate significant improvements in other reasoning and science tasks. AI
IMPACT Provides a scalable, open-source environment for training and evaluating AI agents in software engineering tasks, potentially accelerating progress in the field.
RANK_REASON The cluster describes a new research paper detailing a novel framework and dataset for training AI agents. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
