New benchmark tests if AI code would pass human code review

Original: Frontier Code

Source: cognition.com ↗