SWE-bench Verified

AI Coding Also Creates Technical Debt: Code Simplicity Needs Its Own Benchmark

Parsimony measures AI-generated code complexity; smarter models do not always write less, and the runtime harness can matter more than the model.