Research map / Large language models / Agents, code and math
Strategy-Aware Optimization Modeling with Reasoning LLMs
Ruiqing Zhao, Fengzhi Li, Yuan Zuo, Rui Liu, YanSong Liu, Yunfei Ma, Fanyu Meng, Junlan Feng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add RL with solver feedback after SFT (ablation) | SAGE · 8-benchmark average | pass@1: 74.0 -> 80.3 | Improved |
| add structured reasoning template in RL (ablation) | SAGE · 8-benchmark average | pass@1: 78.3 -> 80.3 | Improved |
| add structured reasoning template in RL (ablation) | SAGE · MAMO Complex | pass@1: 77.5 -> 84.7 | Improved |
| add segment-weighted GRPO (vs uniform weights) (ablation) | SAGE · 8-benchmark average | pass@1: 76.7 -> 80.3 | Improved |
| add segment-weighted GRPO (vs uniform weights) (ablation) | SAGE · MAMO Complex | pass@1: 73.9 -> 84.7 | Improved |
| add efficiency reward (ablation) | SAGE · 8-benchmark average | pass@1: 77.4 -> 80.3 | Improved |
| add efficiency reward (ablation) | SAGE · MAMO Easy | pass@1: 95.0 -> 94.7 | No clear effect |
| add structured template in RL (ablation) | SAGE · IndustryOR Hard subset | accuracy: 33.3 -> 50.0 | Improved |
| add structured template in RL (ablation) | SAGE · IndustryOR Easy subset | accuracy: 81.8 -> 77.3 | Hurt |
| use SAGE-DS-14B instead of SIRL-Q2.5-7B | SAGE · 8-benchmark average | pass@1: 72.7 -> 80.3 | Improved |
| use SAGE-Qwen3-8B instead of SIRL-Q2.5-7B | SAGE with Qwen3-8B backbone · 8-benchmark average | pass@1: 72.7 -> 77.0 | Improved |
| use SAGE instead of SIRL for generation diversity | SAGE · Pass@16 variable types | distinct variable types: 1.80 -> 2.33 | Improved |
Closest papers by meaning
- PEARL: Solver-in-the-Loop Interactive Optimization Modeling from Natural Language (NeurIPS 2026 · similarity 0.88)
- Solver-Informed RL: Grounding Large Language Models for Authentic Optimization Modeling (NeurIPS 2025 · similarity 0.87)
- MM-OptBench: A Solver-Grounded Benchmark for Multimodal Optimization Modeling (NeurIPS 2026 · similarity 0.87)
- REASONING COMPILER: LLM-Guided Optimizations for Efficient Model Serving (NeurIPS 2025 · similarity 0.86)
- OptMATH: A Scalable Bidirectional Data Synthesis Framework for Optimization Modeling (ICML 2025 · similarity 0.86)
- SolverLLM: Leveraging Test-Time Scaling for Optimization Problem via LLM-Guided Search (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
