Error when attaching images via Assistants API

API Bugs
1

Environment

I am communicating with the Assistants API via a Next.js route handler using an OpenAI client instanced from "openai": "^4.103.0" package.

The assistant is using GPT-4o, but the same issue was encountered when trying GPT-4.1.

Steps

  1. User uploads a jpeg image with purpose: vision.
  2. Image gets uploaded successfully; It is visible in the OpenAI dashboard’s storage panel.
  3. User sends a message with said image’s file_id attached in the following format:
{
  "assistantId": "asst_{id}",
  "threadId": "thread_{id}",
  "message": [{
    "type": "text",
    "text": "What do you see?"
  },
  {
    "type": "image_file",
    "image_file": {
      "file_id": "file-{id}"
    }
  }]
}
  1. The API responds with
    { "threadId": "thread_{id}", "messageId": "msg_{id}" }
    without any assistant response.

Alternative flow

We also have an alternative way of communication via WhatsApp. For this, we call the Assistants API directly. When an image is appended, we only get the following error as response: server_error: Sorry, something went wrong.

3 Likes
2

We are getting the same issue. Text works fine, but if an image is attached to request we get the something went wrong error.

This is both in the playground and via API call. I am not sure how to get help for this issue, it’s been almost a 24 hour outage and we’ve had to implement a backup method

3

Thanks for sharing. We had to disable the image attachment feature entirely while this lasts.

What backup method did you find useful?

4

We already had a backup method partially added for cases when OpenAI went down, for service continuity reasons. We just extended this so now we’re operating exclusively on our backup method. We plan to expand it even more so we’re covered in all cases if multiple providers go down.

I really liked using Assistants here, and am sad to see them go. But, since Assistants are being phased out anyway it might be time to move on sooner than later.

Hopefully someone from OpenAI will respond here and fix the issue, but I’m guessing it’s not a priority

1 Like
5

We are getting the same errors, via API and when testing in developer. Assistants are supposed to be online for a few more months, so it needs to be fixed.

I’d assume other companies like ours are still planning and scheduling the migration to the new methods required in the future.

anyone from OpenAI can chime in here on a fix?

6

I’m experiencing the same issue in my app, and I also received a message from the OpenAI incident system about “High errors with image generation.”
This morning, I received another message stating that the issue was “resolved,” but it’s still not resolved for me.
Runs always fail when there are images in the message list.

7

Thanks for raising this!
I was able to reproduce the issue and have flagged it to the team.

3 Likes
8

Thank you @vb please let us know if you hear if/when it’s been fixed. I’ve been reaching out to support, but I’m not sure they’ve escalated the issue

1 Like
9

Thanks @vb

We’re experiencing the issue as well and threads are crashing when they contain an image. Hopefully it’s solved soon.

1 Like
10

Same here - it started on Feb 11 around 2PM PT

1 Like
11

We are having the same issue since February 11th

13

So this has been down for a few days now, has anyone else ever had this kind of experience with the OpenAI api being unreliable (not fixing things)? Makes me question how future developments will play out when they update, etc…

14

Same here, still not working… playground as well

15

132016
1320161920×3413 712 KB

16

What happens next?
The function hasn’t been working for days. Should we switch to Claude? Claude doesn’t just switch off functions.

image
image3587×1693 219 KB

17

Hi @vb Any news? I just went through 3 days of support saying the issue was probably on my side, and now they say they’ll escallate to the engineering team, but it will probably take a couple of days for them to respond to get more information from me (???).

It seems like no one is investigating this at all on OpenAI’s side, we need a little assurance here that someone is really looking into this issue.

1 Like
18

I pinged the team again earlier today already but have no timeline to share.

19

Thank you - We hope to get an update ASAP

20

Hello, we face this issue for days now and I would like to ask if anyone have found a workaround by using other model versions while gpt-4.1 have these issues.

21

Still nothing, our team already started looking to migrate to some other models with vision capabilities, there’s more options here in 2026