OPT-Engine: Benchmarking the Limits of LLMs in Optimization Modeling via Complexity Scaling

Overview

We introduce OPT-Engine, an extensible benchmark framework featuring quantifiable and controllable complexity. OPT-Engine spans ten canonical operations research problems, systematically progressing from Linear Programming to Mixed-Integer Programming. This structured hierarchy provides a principled environment for evaluating automated problem formulation and solving techniques.

The framework currently includes two core problem families:

1. Linear Programming (LP) – Inventory Problem ; Portfolio Allocation Problem; Production Problem; Transportation Problem; Pollution Control Problem;
2. Mixed-Integer Programming (MIP) – : Traveling Salesman Problem (TSP); Knapsack Problem; Bin Packing Problem; Job-Shop Scheduling Problem; Minimum-Cost Network Flow Problem

By offering scalable instances across these fundamental problem types, OPT-Engine enables systematic assessment of solver performance and formulation robustness across varying levels of computational complexity.

Updates

• 2026.01.29 - OPT-Engine paper published on arXiv: OPT-Engine: Benchmarking the Limits of LLMs in Optimization Modeling via Complexity Scaling.

The OPT-Engine Pipeline

Static Benchmark Dataset

Pure-text Reasoning vs. Tool-integrated Reasoning

Identify The Primary Bottleneck