Issue Resolution on SWE-Bench Sphinx Verified (test)
43.51AccuracyGLM-4.5-Air
Evaluation Results
| Method | Links | |
|---|---|---|
| GLM-4.5-AirContext window=32K, Role=Teacher2026.01 | 43.51 | |
| Devstral-Small-2-24BContext window=32K2026.01 | 38.95 | |
| SERA-32B-SphinxContext window=32K, Base Model=Qwen 3-32B, Training trajectories=8,000 synthetic, Repository Specialization=Sphinx2026.01 | 37.14 |