{"id":"arxiv/2610.38175","url":"https://arxiv.org/abs/2610.38175","tags":["cs.AI","math.OC"],"score":0,"title":"DecompGym: Benchmarking LLM Agents on Designing Decomposition Algorithms","domain":"cs.AI","source":"arxiv","authors":["Sikai Cheng","Tinghan Ye","Arnaud Deza","Xiaonan Qi","Miao Li","Litong Liu","Changkun Guan"],"reviews":0,"subject":"cs.AI","summary":"Benders decomposition and column generation solve large optimization problems by passing cuts, columns, and dual values between a master problem and subproblems, and are hard to implement correctly. Large language models now automate parts of algorithm design, which points toward agents that build decomposition solvers for new problems. As a first step toward such agents, we study how well current coding agents design and implement these two algorithms, and how they fail. We introduce DecompGym, 81 instances from nine problem families (five Benders, four column generation), in which an agent…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-10-28","reactions":{}}