After the client version was upgraded to 1.19.1, claude-opus-4.5, which originally supported image understanding, now needs to call the OCR tool to recognize images. Is this a bug or has it adjusted the ability to access the API?