Aider Code Editing
Code
Aider's code-editing benchmark asks a model to complete 133 Exercism Python exercises by editing the provided source files from natural-language instructions, with a second attempt allowed after seeing failing test output. The score is the percentage of exercises whose tests pass. It is the predecessor of the 225-exercise Aider Polyglot benchmark; edit format is a run-level harness setting. Higher is better.