Research map / Large language models / Agents, code and math
OR-R1: Automating Modeling and Solving of Operations Research Optimization Problem via Test-Time Reinforcement Learning
Zezhen Ding, Zhen Tan, Jiheng Zhang, Tianlong Chen
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 TGRPO on unlabeled test data after SFT(3K) (ablation) | Qwen3-8B · OR benchmarks (average) | solution accuracy (%): 64.6 -> 67.7 | Improved |
| add TGRPO on unlabeled test data after SFT(100) (ablation) | Qwen3-8B · OR benchmarks (average) | solution accuracy (%): 59.5 -> 65.9 | Improved |
| use Qwen3-8B instead of Qwen2.5-7B or Llama3-8B as SFT base (ablation) | SFT(3K) · OR benchmarks (average) | solution accuracy (%): 58.2 -> 64.6 | Improved |
| add format reward alone (ablation) | TGRPO · OR benchmarks (average) | solution accuracy (%): 66.0 -> 66.5 | Improved |
| add valid-code reward alone (ablation) | TGRPO · OR benchmarks (average) | solution accuracy (%): 66.0 -> 67.2 | Improved |
| add majority voting reward alone (ablation) | TGRPO · OR benchmarks (average) | solution accuracy (%): 66.0 -> 68.0 | Improved |
| combine all three rewards (ablation) | TGRPO · OR benchmarks (average) | solution accuracy (%): 66.0 -> 70.8 | Improved |
| add format reward on top of code and voting rewards (ablation) | TGRPO · OR benchmarks (average) | solution accuracy (%): 69.7 -> 70.8 | Improved |
| add RL rewards (full) on MAMO ComplexLP (ablation) | TGRPO · MAMO ComplexLP | solution accuracy (%): 42.3 -> 66.8 | Improved |
| add RL rewards (full) on ComplexOR (ablation) | TGRPO · ComplexOR | solution accuracy (%): 44.4 -> 50.0 | Improved |
| add Rcode alone on ComplexOR (ablation) | TGRPO · ComplexOR | solution accuracy (%): 44.4 -> 38.9 | Hurt |
| add TGRPO training data from 10 to 50 samples per test set (ablation) | TGRPO · OR benchmarks (average) | solution accuracy (%): 66.0 -> 69.1 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- LLMOPT — LLMOPT: Learning to Define and Solve General Optimization Problems from Scratch (ICLR 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Solver-Informed RL: Grounding Large Language Models for Authentic Optimization Modeling (NeurIPS 2025 · similarity 0.90)
- StepORLM: A Self-Evolving Framework With Generative Process Supervision For Operations Research Language Models (ICLR 2026 · similarity 0.89)
- Autoformulation of Mathematical Optimization Models Using LLMs (ICML 2025 · similarity 0.89)
- PEARL: Solver-in-the-Loop Interactive Optimization Modeling from Natural Language (NeurIPS 2026 · similarity 0.89)
- Step-by-Step Optimization-like Reasoning in LLMs over Expanding Search Spaces (NeurIPS 2026 · similarity 0.88)
- Large Language Models as End-to-end Combinatorial Optimization Solvers (NeurIPS 2025 · similarity 0.88)
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.
