“The Frontier Code benchmark evaluates AI models not just on passing unit tests, but also on code quality metrics such as scope, discipline, style, and adherence to standards.”