Minwei Kong

A benchmark of 180 literature-grounded OR tasks, each packaged as a mathematical formulation, reference Gurobi implementation, test instances, reference solutions, and an automated feasibility checker. Designed for evaluating LLMs on the end-to-end task of turning a research paper's OR problem into runnable, verifiably-correct optimization code. This repository now contains 180 canonical tasks under tasks/. The legacy ostrowski2012 compatibility alias, which duplicated earl2005, was removed on 2026-09-06. For the per-paper instance / solution data, clone or download the full repository — the structure is intentionally a file tree, not a flat HF dataset, because each paper is its own…