FrontierCode

The article introduces FrontierCode, a new benchmark for evaluating the ability of AI models to write high-quality, maintainable code that would be accepted by human maintainers in production codebases.