Software Engineering Code Generation on SWE-bench
65Test Success Rate (TSR)OpenCode (via ADD S.)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| OpenCode (via ADD S.)Injection Strategy=via ADD S., Backend Model=Qwen 3.6 Plus2026.05 | 65 | 95 | 95 | 95 | |
| Claude Code (via EDIT S.)Injection Strategy=via EDIT S., Backend Model=Qwen 3.6 Plus2026.05 | 60 | 90 | 90 | 10 | |
| Kilo CodeInjection Strategy=None, Backend Model=Qwen 3.6 Plus2026.05 | 55 | — | — | — | |
| Kilo Code (via EDIT S.)Injection Strategy=via EDIT S., Backend Model=Qwen 3.6 Plus2026.05 | 55 | 60 | 60 | 30 | |
| Claude CodeInjection Strategy=None, Backend Model=Qwen 3.6 Plus2026.05 | 50 | — | — | — | |
| OpenCodeInjection Strategy=None, Backend Model=Qwen 3.6 Plus2026.05 | 50 | — | — | — | |
| OpenCode (via EDIT S.)Injection Strategy=via EDIT S., Backend Model=Qwen 3.6 Plus2026.05 | 50 | 80 | 80 | 50 | |
| Kilo Code (via ADD S.)Injection Strategy=via ADD S., Backend Model=Qwen 3.6 Plus2026.05 | 45 | 85 | 85 | 85 | |
| Claude Code (via ADD S.)Injection Strategy=via ADD S., Backend Model=Qwen 3.6 Plus2026.05 | 40 | 20 | 20 | 15 |