Data Processing Workflow Generation on DataGovbench DAG-Level
28.65ATSDeepSeek-V3
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DeepSeek-V32025.12 | 28.65 | 56 | 72 | 52.22 | 983.7 | 1,098.9 | 305.99 | |
| Qwen3-235b-a22b2025.12 | 25.64 | 38 | 50 | 37.88 | 3,005.22 | 7,339.2 | 81.43 | |
| Qwen3-coder2025.12 | 20.87 | 36 | 48 | 34.96 | 1,075.36 | 77.32 | 370.27 | |
| Qwen2.5-coder2025.12 | 12.11 | 26 | 30 | 22.7 | 738.68 | 852.36 | 28.23 | |
| Gemma-3-27B2025.12 | 11.31 | 20 | 38 | 23.1 | 1,671.34 | 2,412.24 | 19.06 | |
| Llama-3-70B2025.12 | 8.07 | 10 | 16 | 11.36 | 723.08 | 284.43 | 221.09 | |
| Llama-4-scout2025.12 | 7.35 | 12 | 22 | 13.78 | 864.16 | 435.08 | 10.39 | |
| Mistral-7B2025.12 | 7.1 | 18 | 20 | 15.03 | 897.88 | 261.9 | 230.13 | |
| Phi-42025.12 | 6.73 | 20 | 28 | 18.24 | 1,081.94 | 929.29 | 18.35 |