500 Internal Server Error with o3‑mini + file_search on EastUS2 region (Assistants API)

Question

500 Internal Server Error with o3‑mini + file_search on EastUS2 region (Assistants API)

Goran 0

Our application has been successfully creating Azure OpenAI assistants with o3-mini (2025‑01‑31 build), attaching a vector store, enabling file_search, and calling /runs on the Assistants API. As of today, these calls now return a 500 Internal Server Error:

POST https://<your‑endpoint>.openai.azure.com/openai/threads/<thread_id>/runs?api‑version=2024‑05‑01‑preview

500 Internal Server Error

{

"error": {

"message": "The server had an error processing your request. Sorry about that! You can retry your req (truncated...)"

}

This flow worked without issue until very recently, with no changes in our code or Azure configuration.

Steps to reproduce:

Create an assistant with model = o3-mini (2025-01-31).

Attach tools including file_search, plus a file-backed vector store.

Creates assistant thread, then POSTs to /runs.

The POST returns an HTTP 500 Internal Server Error (no further details).

[EDIT]: It looks like error is happening when passing reasoning_effort paramenter when creating assistant. Per OpenAI docs that is possible for o-series models only. I omitted and passed that argument when calling assistant run and seem to work now.

Manas R Mohanty 17,275 Reputation points Moderator

2025-07-07T09:19:22.99+00:00

Hi Goran

Good day. Hope you found the answer from Amira Bedhiafi helpful.

Please let us know if omitting reasoning effort param from assistant calls helped resolve the isue.

Thank you
Goran 0 Reputation points

2025-07-07T09:35:50.97+00:00

Hi Amira

interestingly enough passing reasoning_effort while calling /runs works. However when creating Assistants only then is it throwing error. Having said that I’m unsure weather it is actually being used when calling /runs with it. But my question is, if there are any new api versions or region that supporta reasoning effort for assistants api with file search?
Goran 0 Reputation points

2025-07-08T11:02:11.6833333+00:00

Thanks for the update! Yes, confirmed even without reasoning effort, it’s working using the 2025-04-01-preview API version and o3-mini in East US. No 500 errors, and file search + vector store setup runs as expected.
Manas R Mohanty 17,275 Reputation points Moderator

2025-07-08T14:46:07.5+00:00

Hey Goran

Could you please accept below workaround as answer then.

Thank you.

3 answers

Your answer

Manas R Mohanty 17,275 Reputation points Moderator

2025-07-07T09:19:22.99+00:00

Hi Goran

Good day. Hope you found the answer from Amira Bedhiafi helpful.

Please let us know if omitting reasoning effort param from assistant calls helped resolve the isue.

Thank you
Goran 0 Reputation points

2025-07-07T09:35:50.97+00:00

Hi Amira

interestingly enough passing reasoning_effort while calling /runs works. However when creating Assistants only then is it throwing error. Having said that I’m unsure weather it is actually being used when calling /runs with it. But my question is, if there are any new api versions or region that supporta reasoning effort for assistants api with file search?
Goran 0 Reputation points

2025-07-08T11:02:11.6833333+00:00

Thanks for the update! Yes, confirmed even without reasoning effort, it’s working using the 2025-04-01-preview API version and o3-mini in East US. No 500 errors, and file search + vector store setup runs as expected.
Manas R Mohanty 17,275 Reputation points Moderator

2025-07-08T14:46:07.5+00:00

Hey Goran

Could you please accept below workaround as answer then.

Thank you.

Answer 1

Deleted

This answer has been deleted due to a violation of our Code of Conduct. The answer was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.

Comments have been turned off. Learn more

Answer 2

Deleted

This answer has been deleted due to a violation of our Code of Conduct. The answer was manually reported or identified through automated detection before action was taken. Please refer to our Code of Conduct for more information.

Comments have been turned off. Learn more

Answer 3

Hello Goran !

Thank you for posting on Microsoft Learn.

What you are sharing is a known limitation introduced in the Assistants API (May 2024 preview) when using the o3-mini model with file_search and reasoning_effort in the EastUS2 region.

You're right in identifying that the problem is caused by passing the reasoning_effort parameter when starting a run (/runs endpoint) with a model (o3-mini) that currently does not support it under Azure Assistants API, despite OpenAI general documentation saying it's allowed for o-series models.

reasoning_effort is only supported for chat completions, not for assistant runs using file search with o3-mini.

You can omit the reasoning_effort parameter when creating the assistant or calling /runs for that assistant.

Even though reasoning_effort is accepted by the OpenAI API, Azure Assistants API might still enforce stricter checks or lag slightly behind in compatibility with model features.

Try to use:

{
  "assistant_id": "...",
  "instructions": "Answer using uploaded files",
  "tools": [{"type": "file_search"}]
}