Agentic Coding on SWE-Bench (Score, Relative Change)
59ScoreSAR
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| SARAgent=ClaudeCode, Projection=top-1% spectral-rank2026.07 | 59 | 1.72 | |
| Baseline (In-house)Agent=ClaudeCode2026.07 | 58 | — | |
| SARAgent=CodeAct, Projection=top-1% spectral-rank2026.07 | 57.4 | 1.41 | |
| Baseline (In-house)Agent=CodeAct2026.07 | 56.6 | — |