This blog post discusses issues with the Code Interpreter in the GPT-4.1 Responses API, particularly the lack of proper file download link generation and poor tool-following due to internal instructions. It highlights a model fault and proposes solutions for improving output consistency.