Multimodal API fails on one valid image URL across all vision-capable models — error unclear

API
1

This is really strange: whenever I send a completions request with a message containing an image_url , no matter which vision-capable model I use (including gpt-4.1-mini, gpt-4o, gpt-5 ), I consistently get the same unhelpful error—no matter how many times I retry the API call.

The problematic image URL

https://ptboxs.obs.cn-east-3.myhuaweicloud.com/images/3a1616ce40eaee72acbf768a5d94741f/f84f5541e97f4683b20b6da6ded1cbd7/a50ceb3304c34cdbb06cd466cd70758e/2026/2/27/K5FWH%7D%60K%407V7HU%5DS9MP13BF.png___3a1fb02a1d4cc783cec54f0a29378824.png?AccessKeyId=CMAQKLU6J3FGLYU2CHKB&Expires=1772211196&response-cache-control=no-cache&Signature=4wBuiOONJ1ENMXbfwVDlsEEGP3s%3D

Note:

  • If I remove the Signature parameter from the URL query string, the request succeeds.
  • This URL may expire in 6 hours.

Errors

"Error code: 400 - {'error': {'message': 'Error while downloading https://ptboxs.obs.cn-east-3.myhuaweicloud.com/images/3a1616ce40eaee72acbf768a5d94741f/f84f5541e97f4683b20b6da6ded1cbd7/a50ceb3304c34cdbb06cd466cd70758e/2026/2/27/K5FWH%7D%60K%407V7HU%5DS9MP13BF.png___3a1fb02a1d4cc783cec54f0a29378824.png?AccessKeyId=CMAQKLU6J3FGLYU2CHKB&Expires=1772211196&response-cache-control=no-cache&Signature=4wBuiOONJ1ENMXbfwVDlsEEGP3s%3D.', 'type': 'invalid_request_error', 'param': None, 'code': 'invalid_image_url'}}"

I don’t understand why the API treats this image_url as invalid when a nearly identical one succeeds. https://ptdatas.obs.cn-east-3.myhuaweicloud.com/files/39f89a2bd7b37dd6f3755193ba020f15/351cd5bf57a34286a08717f398cdb309/c63936b5797249c693ccb4cd814861ea/2026/1/5/62f65174327ffa0002aba557.jpg___3a1e9feef6dd584febddacd1f98015a9.jpg?AccessKeyId=CMAQKLU6J3FGLYU2CHKB&Expires=1772206794&response-cache-control=no-cache&Signature=raBUBigpogikUCya3IY%2B%2BPlWKTI%3D

my code:

response = client.chat.completions.create(
    model="gpt-4.1-mini",
    messages=[
        {
            "role": "user",
            "content":
                [  
                {"type":"text","text":"What did you see？"},
                {"type":"image_url","image_url":{"url":"https://ptboxs.obs.cn-east-3.myhuaweicloud.com/images/3a1616ce40eaee72acbf768a5d94741f/f84f5541e97f4683b20b6da6ded1cbd7/a50ceb3304c34cdbb06cd466cd70758e/2026/2/27/K5FWH%7D%60K%407V7HU%5DS9MP13BF.png___3a1fb02a1d4cc783cec54f0a29378824.png?AccessKeyId=CMAQKLU6J3FGLYU2CHKB&Expires=1772211196&response-cache-control=no-cache&Signature=4wBuiOONJ1ENMXbfwVDlsEEGP3s%3D"}}

                ]
        }],
    temperature=0.7,
    max_tokens=1000
)

I can access the image URL directly in my browser without any issue, and models like Gemini can correctly understand the image content. It almost seems as if OpenAI is modifying or rejecting this specific URL—but I have no idea why. So far, this problem only occurs with this particular image URL.

Could anyone explain what’s going on? Without understanding the root cause, we’re left completely in the dark—we won’t know when or why it might happen again, and we can’t proactively avoid it.