Graph Reasoning on GRbench
48.87E-COMMERCE (R-L) ScoreGRAPHDANCER
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| GRAPHDANCERModel=Qwen2.5-3B-Instruct2026.01 | 48.87 | 47 | 43.82 | 49.58 | 31.55 | 29.63 | 38.25 | 42.78 | 40.62 | 42.25 | |
| Vanilla RLModel=Qwen2.5-3B-Instruct2026.01 | 45.66 | 45 | 47.36 | 49.17 | 27.09 | 25.93 | 33.84 | 36.11 | 38.49 | 39.05 | |
| Graph-CoTModel=Qwen2.5-3B-Instruct2026.01 | 43.58 | 41 | 43.6 | 47.5 | 27.47 | 26.67 | 30.68 | 35.56 | 36.33 | 37.68 | |
| Graph-CoTModel=GPT-3.5-turbo2026.01 | 42.4 | 44.5 | 41.59 | 46.25 | 22.33 | 29.89 | 30.52 | 28.33 | 34.21 | 37.24 | |
| Graph-CoTModel=Qwen3-14B2026.01 | 39.91 | 42.5 | 48.03 | 53.33 | 38.82 | 36.67 | 32.66 | 35.56 | 39.85 | 42.01 | |
| Graph-CoTModel=GPT-4o-mini2026.01 | 37.06 | 39.5 | 36.04 | 46.25 | 39.88 | 41.48 | 35.47 | 40 | 37.11 | 41.81 | |
| ToG-2Model=GPT-4o-mini2026.01 | 29.28 | 35 | 21.97 | 30.42 | 18.66 | 18.89 | 25.59 | 23.33 | 23.88 | 26.91 | |
| GraphRAGModel=GPT-3.5-turbo2026.01 | 17.52 | 28 | 14.94 | 24.17 | 8.69 | 14.07 | 18.66 | 22.22 | 14.95 | 22.12 | |
| TextRAGModel=GPT-3.5-turbo2026.01 | 14.06 | 20 | 10.04 | 20.83 | 4.57 | 8.52 | 18.14 | 23.89 | 11.7 | 18.31 |