Extra Task Completion on GAIA
87.1Extra Task Completion RateDeepSeek (Baseline)
Evaluation Results
| Method | Links | |
|---|---|---|
| DeepSeek (Baseline)Model=DeepSeek2025.12 | 87.1 | |
| LeechHijackModel=DeepSeek, Agent Framework=Hybrid (Pydantic-AI)2025.12 | 86.84 | |
| LeechHijackModel=DeepSeek, Agent Framework=Cloud-Hosted (OpenManus)2025.12 | 79.56 | |
| LeechHijackModel=DeepSeek, Agent Framework=Localized (OpenManus)2025.12 | 67.91 | |
| BaselineBackbone Model=GPT2025.12 | 0.887 | |
| BaselineBackbone Model=Deepseek2025.12 | 0.871 | |
| BaselineBackbone Model=Gemini2025.12 | 0.834 | |
| BaselineBackbone Model=Qwen2025.12 | 0.8138 | |
| LeechHijackBackbone Model=GPT2025.12 | 0.7561 | |
| LeechHijackBackbone Model=Deepseek2025.12 | 0.6791 | |
| LeechHijackBackbone Model=Qwen2025.12 | 0.65 | |
| LeechHijackBackbone Model=Gemini2025.12 | 0.4362 |