arXiv cs.AIWednesday · May 27, 2026FREE

Generating Robust Portfolios of Optimization Models using Large Language Models

A new paper on arXiv (2605.27013) introduces a method for generating robust portfolios of optimization models using large language models. The algorithm leverages a single LLM in two roles: as a stochastic generator to produce multiple candidate models, and as a reasoning evaluator to assess them. This portfolio approach mitigates the risk of relying on any single LLM-generated model, which may be unreliable. The authors provide theoretical guarantees that the portfolio is robust as long as either the generator or evaluator performs adequately. The work addresses the bottleneck of formulating optimization models, which typically requires both domain expertise and optimization knowledge. By using natural language descriptions, the method could make optimization more accessible to non-experts. The paper is published on arXiv and was announced on May 27, 2026.