A Single Untuned Prompt Gets GPT-5.6-Sol to Write OR Algorithms That Match the Best Known Methods

arXiv 2608.27296·medium signalXBlueskyLinkedInCopy link

The study tests inventory control, queueing network control and assortment optimization at two levels: solving one given instance, and returning an algorithm that maps instance parameters to solutions before seeing any evaluation instance. With one untuned prompt describing the problem and a Python sandbox on a fixed compute budget, gpt-5.6-sol matches or beats the best existing method on almost all evaluated instances, including at the harder level-2 setting. Performance improves sharply across models released less than eight months apart, which the authors read as reason to treat a frontier LLM as a serious empirical baseline for algorithm design in well-specified OR problems.