Assistant runs with file uploads failing

Summary

Since November 10 2025, all Assistant API runs that include uploaded files have started failing with the message:

**“Run failed – Sorry, something went wrong.”

Runs without files continue to succeed.


Impact

  • Affects all assistants and environments (production and development).

  • Blocks any workflow relying on file uploads (e.g., file search store attachments).

  • Breaks previously working code with no recent code or configuration changes on our side.


Reproduction Steps

  1. Upload a file (tested with both .pdf ~50 KB and .txt ~11 KB).

  2. Create a thread referencing the uploaded file.

  3. Start a run (POST /v1/threads/{thread_id}/runs).

  4. The run fails immediately with the generic error above (shows as the content of the last error field).

When the same content is sent inline in the message body instead of as an uploaded file, the run succeeds.


Observed Behavior

  • Run status shows as failed with the last error showing only the generic error message

  • Dashboard shows file search stores attached to the thread, with reasonable sizes, but runs still fail.

  • Verified that these same files and workflow worked successfully prior to today.


Expected Behavior

Assistant runs with attached files should execute normally, as they did previously, without file-related errors.


Environment

  • Model: gpt-4o

  • API: OpenAI Assistants API

  • File upload purpose: assistants

  • Files tested: .txt, .pdf (≤ 50 KB each)

  • Reproduced on: multiple environments, identical behavior

2 Likes

Easy reproduction in the platform site:

A gpt-4.1 Assistant with a vector store, informed in the system message “Search OpenAI knowledge in provided files with tools before answering.” Doesn’t call tools and can’t answer about Responses correctly, fabricating docs it “read”.

Attach 2MB text in the user input box as vector store file, results in Error instead of another response:

1 Like

Yes to follow up with what _j reported, this can also be seen from the assistants playground on the platform website.

I see in the browser network tab that submitting a message with an attachment returns a 500 response code from this url: https://api.openai.com/v1/threads/thread_dVkP9YErY66zrAb60LTaHoXB/messages

{“role”:“user”,“content”:[{“type”:“text”,“text”:“Please summarize what you see in the attached document.”}],“attachments”:[{“file_id”:“”,“tools”:[{“type”:“file_search”}]}]}

It’s now been 24 hours and the status page still doesn’t show any issues having been tracked or fixed. This issue has us largely dead in the water with regards to processing uploaded files.

This really reduces confidence when there is clearly an issue, but the status page still reports everything being 100% functional.

I’m running into the exact same issue. No workarounds that I’ve tried seem to help.

This is now working for our application again (Assistant API file uploads), at least intermittently. When I attempt to upload files in the platform ui, I still get the same error, but that is of lesser concern to me at this point.

FWIW this has sparked us to accelerate the move to Responses API. Hoping that a non-deprecated feature will see greater reliability.

Getting the same issue here

It seems to be isolated to the reasoning models (o1 and o3-mini). Those fail no matter what setting you use