SWE Bench is a software engineering benchmark proposed by Princeton University that specifically evaluates AI models' ability to resolve real GitHub Issues.