Inference Time Consumption on GPQA
14.7Inference Time: Research And DataNormal
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| NormalModel=Gemini2025.12 | 14.7 | 96.3 | 31.2 | — | 31.3 | |
| LeechHijackModel=Gemini2025.12 | 16.4 | 71.3 | 24.4 | — | 22 | |
| NormalModel=Qwen2025.12 | 22 | 36.6 | 20 | — | 19.8 | |
| NormalModel=GPT2025.12 | 29.1 | 52.3 | 20.1 | — | 20.6 | |
| LeechHijackModel=GPT2025.12 | 30.5 | 58.5 | 23.4 | — | 24.1 | |
| LeechHijackModel=Qwen2025.12 | 34.5 | 53.6 | 28.1 | — | 27.7 | |
| Hybrid (Pydantic-AI)MCP Tool=Normal2025.12 | 53.2 | 74.5 | 47 | — | 47.1 | |
| Hybrid (Pydantic-AI)MCP Tool=LeechHijack2025.12 | 58 | 99.5 | 54.5 | — | 54.5 | |
| Localized (OpenManus)MCP Tool=LeechHijack2025.12 | 69.8 | 145 | 120.4 | — | 96.9 | |
| LeechHijackModel=Deepseek2025.12 | 69.8 | 145 | 120.4 | — | 96.9 | |
| Cloud-Hosted (OpenManus)MCP Tool=Normal2025.12 | 72.7 | 122.2 | 79.6 | — | 79.6 | |
| Localized (OpenManus)MCP Tool=Normal2025.12 | 76.2 | 139.2 | 91.1 | — | 101.5 | |
| NormalModel=Deepseek2025.12 | 76.2 | 139.2 | 91.1 | — | 101.5 | |
| Cloud-Hosted (OpenManus)MCP Tool=LeechHijack2025.12 | 81.4 | 135.1 | 91.8 | — | 91.8 | |
| Hybrid (OWL)MCP Tool=Normal2025.12 | 208.1 | 349 | 295.5 | — | 325.5 | |
| Hybrid (OWL)MCP Tool=LeechHijack2025.12 | 210.7 | 269.1 | 269.4 | — | 253 |