Benchmarking Cline w/ updated prompt on SWE-Bench Lite

GPT-4.1

Claude Sonnet 4.5

~15%

Improvement, just through rules

No fine-tuning, no tool changes, no architecture changes. JUST RULES.
GPT-4.1 achieved performance near Sonnet 4.5, which is widely considered state of the art for coding questions
2/3 cost!