Graphs in assistants not working

Hi,

I’m using the assistants in both api and the playground, suddenly the matplot graphs stopped working.

"last_error":{
"code":"server_error"
"message":"Sorry, something went wrong."
}

This is my prompt
Create a graph showing the price of the dollar in 2023 vs mexican peso using random values

This start happening a few moments ago. I’m using gpt-4o. Is anyone else having this problem?

2 Likes

Experienced the same. Charts are not being generated with the assistants API.

I also get an error when uploading a file or image.

Same with me too for the images and chat. Can someone guide us

Hello,

I don’t know you issue but on my side if I upload CSV file and ask the assistnat to generate a graph using the data it works fine!

Can you share more information about your issue so we can help?

I also use CSVs to generate graphs and charts, so I created a very simple assistant, and asked to create a random graph, here are two images. In the first one the code interpreter starts working as usual, the problem appear to be the plt.show()

Because then I got this

Same issue being discussed in this thread: Run failed, Sorry, something went wrong. Error while sending an image

2 Likes

Im using an external tool and it is working but when i try it in the openai console I also have the issue


So im able to make it work here but I cant really find where it is not working in OpenAI becauase im using their API.

Edit: nvm im able to see it now

I have no clue about your issue, im currently using the code execution tool

Edit 2: it works with graph

I have noticed that it has worked a few times, both in our production application, and in the playground. So there is some element of intermittency that is happening.

1 Like

Yeah, I dont know what could be the issue here!

A “hacky” solution was to instruct the agent to generate the graph inside a pdf, that seems to work

2 Likes

It seems to only fail if you upload an image (using file_id). I provided the same image, but via a URL, and vision worked.

Annoying that there is still no word on this, but at least there is a workaround for now.