Models such as GPT-4 and Claude 3 have achieved performance on code generation tasks that approaches or even surpasses the level of mid-level developers.