Autonomous Agentic Data Engineering on Autonomous Agentic Data Engineering Science
7.13MATSDeepSeek-V3.1
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DeepSeek-V3.1Agent Mode=Iterative Agent, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 7.13 | 45.91 | |
| Claude-4-SonnetAgent Mode=One-Shot, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 3 | 30.17 | |
| GPT-5.2Agent Mode=Iterative Agent, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 3 | 82.23 | |
| DeepSeek-R1Agent Mode=Iterative Agent, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2.75 | 89.76 | |
| Claude-4-SonnetAgent Mode=Iterative Agent, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2.61 | 58.93 | |
| DeepSeek-R1Agent Mode=One-Shot, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2.5 | 47.94 | |
| DeepSeek-V3.1Agent Mode=One-Shot, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2.5 | 18.04 | |
| DeepSeek-V3.1Agent Mode=One-Shot, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2.5 | 52.78 | |
| Claude-4-SonnetAgent Mode=One-Shot, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2.5 | 79.48 | |
| Claude-4-SonnetAgent Mode=Iterative Agent, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2.17 | 82.08 | |
| DeepSeek-R1Agent Mode=Iterative Agent, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2.13 | 37.01 | |
| GPT-5.2Agent Mode=One-Shot, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2 | 35.66 | |
| Gemini-2.5-ProAgent Mode=One-Shot, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2 | 35.63 | |
| Qwen3-MaxAgent Mode=One-Shot, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 2 | 49.34 | |
| Qwen3-MaxAgent Mode=Iterative Agent, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1.75 | 74.67 | |
| Gemini-2.5-ProAgent Mode=Iterative Agent, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1.63 | 46.62 | |
| Gemini-2.5-ProAgent Mode=Iterative Agent, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1.6 | 41.79 | |
| Qwen3-MaxAgent Mode=Iterative Agent, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1.58 | 58.24 | |
| GPT-5.2Agent Mode=Iterative Agent, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1.53 | 70.58 | |
| DeepSeek-V3.1Agent Mode=Iterative Agent, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1.47 | 78.14 | |
| Qwen3-MaxAgent Mode=One-Shot, Data Initialization Setting=From Scratch, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1 | 37.69 | |
| GPT-5.2Agent Mode=One-Shot, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1 | 58.24 | |
| DeepSeek-R1Agent Mode=One-Shot, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1 | 83.6 | |
| Gemini-2.5-ProAgent Mode=One-Shot, Data Initialization Setting=With Seed, Teacher Model=Qwen3-30B-A3B, Student Model=Llama-3.1-8B-Instruct2026.05 | 1 | 87.01 |