Software Engineering on SWE-bench (Productivity, Proactivity, Personalization)
21.4ProductivityMCPE
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MCPEAgent=DeepSeek-V3.2, User=Qwen3-VL-32B, Communication Mode=CPE-optimized2026.06 | 21.4 | 7.6 | 58.9 | |
| MhybridAgent=DeepSeek-V3.2, User=Qwen3-VL-32B, Communication Mode=Hybrid2026.06 | 14 | 8.9 | 59.4 | |
| MtextAgent=DeepSeek-V3.2, User=Qwen3-VL-32B, Communication Mode=Text-only2026.06 | 13.5 | 15.6 | 62.9 | |
| MuiAgent=DeepSeek-V3.2, User=Qwen3-VL-32B, Communication Mode=UI-only2026.06 | 11.3 | 19.6 | 70.5 |