As of today, when calling the Assistants API with a query that generates an image, we started receiving the error “server_error” description “Sorry, something went wrong”. No other details. I see in the OpenAI dashboard that the image files are being created.
Attempting to work around the error, and seeing the new project-based keys, I created a project with a new full-access key. Using that key, I receive a new error: “image_file_not_found” description “Failed to fetch image file: file-xxx. Check if the file has been deleted.” And there are no “assistant_output” files listed in storage for the project.
We were using GPT-4o without issue until now. Switching to GPT4-turbo results in the same errors.
3 Likes
I got the same error I’m using GPT-4o-2024-05-13, it appear to be related to the matplotlib, if I gave the instruction to not use that them it can generate the chart
1 Like
Hopefully this indicates they’ve finally realized there’s a problem:
Identified - We have identified the issue causing the increased latency and error rates. We are currently in the process of implementing a fix.
May 23, 2024 - 16:29 PDT
Investigating - We’re currently investigating an elevated level of errors affecting API and ChatGPT users.
May 23, 2024 - 15:34 PDT
per status dot openai dot com
Unfortunately the status page says the issue was resolved but I’m still having this problem.
I have a same error too bro.
Run failed
Sorry, something went wrong.
My project cannot use Image analyzing now 
I hope that this problem will be fixed sooner.
This is GPT-4-Turbo and GPT-4o specific. 3.5-Turbo works fine.
Other threads have suggested this is a new problem with matplotlib, and instructing not to use matplotlib resolves the issue. Haven’t tested yet.
We are facing the same issue.
In our case, we may be able to overcome by retrying with additional instruction or prompt such as “DO NOT USE matplotlib.pyplot, but use matplotlib.figure.Figure” if it happens. The point is that temperature should be low to avoid the randomness. Thanks for your good advice.