DecompGym: Benchmarking LLM Agents on Designing Decomposition Algorithms
This work introduces DecompGym, 81 instances from nine problem families (five Benders, four column generation), in which an agent writes a solver for a prescribed method, and evaluates 12 GPT, Claude, and Gemini models running in coding-agent tools.
Si-Kai Cheng, Ting-Han Ye, Arnaud Deza et al.
· 0 citations