GPT-4v failed in simple text recognition

GPT4v works well in many vision tasks. However, it failed in simple text recognition.
I tried scaling up or down the text in the image, but still didn’t succeed.
It’s interesting that when I used the same prompt to recognize text in different images, it worked for complex scenes but still failed for simple ones.
Even after adding random noise to the text in simple scene images, it still failed to recognize the text.