VRP optimization-modeling on VRP Full 700 problem set
95Pass@1Gemini-3.1-Pro Preview
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3.1-Pro PreviewModel type=Closed-source frontier LLM2026.06 | 95 | |
| VRPCoder-GRPOModel scale=8B, Stage=GRPO2026.06 | 93 | |
| VRPCoder-SFTModel scale=8B, Stage=SFT2026.06 | 88.43 | |
| Claude-Sonnet-4.5Model type=Closed-source frontier LLM2026.06 | 64.71 | |
| DeepSeek-V3.2Model type=Open-source general-purpose LLM2026.06 | 36.29 | |
| MiniMax-M2.5Model type=Open-source general-purpose LLM2026.06 | 24 | |
| Qwen2.5-72B-InstructModel type=Open-source general-purpose LLM2026.06 | 18.57 | |
| SIRL-Gurobi-32BModel type=OR-LLM2026.06 | 15 | |
| ORLM-LLaMA-3-8BModel type=OR-LLM2026.06 | 5.71 | |
| SIRL-Gurobi-8BModel type=OR-LLM2026.06 | 5.57 | |
| OptMATH-Qwen2.5-7BModel type=OR-LLM2026.06 | 3.71 | |
| Qwen3-8BModel type=Open-source general-purpose LLM2026.06 | 0.57 |