I figure a big part of it is that SWE-Bench is the target benchmark for programming and it's all python.
Python being the language LLMs are best at predates SWE-Bench by years.