[2026-04-29T17:36][v2_1736_0429] Claude Opus 4.7 SWE-bench 87.6% vs GPT-5.5 Terminal-Bench 82.7%：代码Agent选型横评