BWOR: A Bilingual Benchmark for Modeling and Solving Operations Research Problems from Natural Language

BWOR is a bilingual benchmark for evaluating systems that transform natural-language operations research (OR) problems into mathematical optimization models, executable solver code, and verified objective values. The benchmark contains 82 textbook-style OR problems with original Chinese statements, normalized English translations, answer labels, solution-status labels, and metadata for domain and mathematical-programming type. This repository is intended as a test-only benchmark. The examples should not be used for training or prompt tuning when reporting benchmark results.