SAC-Opt: Semantic Anchors for Iterative Correction in Optimization Modeling

Researchers have proposed two new frameworks, SAC-Opt and MIRROR, to improve optimization modeling using large language models (LLMs). SAC-Opt enhances modeling accuracy by correcting semantic errors, while MIRROR translates natural language problems into mathematical models and solver code.

SAC-Opt, introduced in a paper submitted to arXiv on 28 September 2025^[1], is a backward-guided correction framework that improves optimization modeling by grounding it in problem semantics. It corrects undetected semantic errors in generated code by aligning original semantic anchors with those reconstructed from the code, enhancing fidelity and robustness without requiring additional training or supervision. According to the authors, SAC-Opt improves average modeling accuracy by 7.7% across seven public datasets, with gains of up to 21.9% on the ComplexLP dataset^[1]. Meanwhile, MIRROR, presented in another arXiv paper^[2], directly translates natural language optimization problems into mathematical models and solver code. MIRROR integrates two core mechanisms: execution-driven iterative adaptive revision and hierarchical retrieval. The framework outperforms existing methods on standard operations research (OR) benchmarks, particularly on complex industrial datasets, providing non-expert users with an efficient and reliable OR modeling solution.