Abstract: Optimization is fundamental to scientific discovery and real-world decision-making, yet translating natural language problem descriptions into mathematical formulations and selecting appropriate solvers remains a challenging task that typically requires substantial domain expertise. We introduce OptimAI, a framework for solving optimization problems described in natural language by leveraging large language model (LLM)-powered AI agents. OptimAI integrates four key components: (1) a formulator that converts natural language into precise mathematical expressions; (2) a planner that outlines high-level solution strategies; and (3) a coder and a code critic that iteratively execute and refine code based on computational feedback. The system is designed to support multi-agent collaboration, enabling the flexible integration of diverse models within a unified framework. OptimAI achieves state-of-the-art performance, attaining 88.1% accuracy on the NLP4LP dataset (a 16.5% improvement) and 71.2% on the nonlinear-without-table subset of Optibench (a 29.1% improvement).