o3-mini Server Error: "Sorry, something went wrong."

Question

o3-mini Server Error: "Sorry, something went wrong."

Andreas Gnyp 10

I get this message constantly:

{"code"=>"server_error", "message"=>"Sorry, something went wrong."}

I'm using o3-mini with Assistant and File Search which takes a vector store (works fine).

Manas Mohanty 17,185 Reputation points Microsoft External Staff Moderator

2025-12-02T06:10:45.93+00:00

Hi Adam Zachary

Hope you have found Adam Zachary's answer useful.

Please let us know if the issue persists for you.

Thank you.
Andreas Gnyp 10 Reputation points

2025-12-02T14:51:01.96+00:00

Hi Manas Mohanty,

unfortunately not. See my comment.

1 answer

Your answer

Manas Mohanty 17,185 Reputation points Microsoft External Staff Moderator

2025-12-02T06:10:45.93+00:00

Hi Adam Zachary

Hope you have found Adam Zachary's answer useful.

Please let us know if the issue persists for you.

Thank you.
Andreas Gnyp 10 Reputation points

2025-12-02T14:51:01.96+00:00

Hi Manas Mohanty,

unfortunately not. See my comment.

Answer 1

Adam Zachary 2,265

I hit the same {"code"=>"server_error", "message"=>"Sorry, something went wrong."} with o3-mini when using Assistants plus file_search on Azure. In my case it was caused by a combination of three things.

Remove reasoning_effort anywhere you use o3-mini with Assistants and file_search Do not set reasoning_effort when you create the assistant, and do not pass it when you create the run. With o3-mini on Azure, reasoning_effort works for normal responses, but it currently breaks when you use Assistants plus file_search and a vector store. As soon as I removed reasoning_effort from both the assistant and the run, the constant server_error stopped.

Update to a recent preview API version and a region where this combination is stable Make sure you are calling a recent preview like 2025-04-01-preview for the Assistants API and that your o3-mini deployment is in a region where Assistants plus file_search is supported, for example East US. When I switched from an older preview version and moved the deployment to a supported region, o3-mini + file_search + vector_store started working consistently.

Avoid forcing file_search unless you really need to If you are setting tool_choice to force file_search, try running first with tool_choice set to auto. If you must force file_search, add simple retry logic around the run call, because you can still get occasional transient server_error responses.

A minimal pattern that works for me now looks like this:

assistant = client.beta.assistants.create(
    model="o3-mini",
    instructions="Answer using the uploaded files.",
    tools=[{"type": "file_search"}],  # no reasoning_effort here
)

assistant = client.beta.assistants.update(
    assistant_id=assistant.id,
    tool_resources={"file_search": {"vector_store_ids": [vector_store.id]}},
)

run = client.beta.threads.runs.create(
    thread_id=thread.id,
    assistant_id=assistant.id,
    # no reasoning_effort here either
    # optionally do not force tool_choice
)

If after removing reasoning_effort, updating the API version, and testing a supported region you still get server_error on every run, capture the x-ms-request-id from the response and open an Azure support ticket, because at that point it is likely a platform issue rather than your code.

Andreas Gnyp 10 Reputation points

2025-12-02T14:40:20.54+00:00

Hi Adam Zachary,

thank you for your quick and extended response.

Your suggestion to go for another region might probably be the best solution. Yet, there is only one reason we're using Azure in the first place: GDPR.

As for the other suggestions, unfortunately they don't change much. This model is constantly flaky. Functioning fine a few days, then it's showing the mentioned server errors, and just now, it works fine with the first thread/run, the next one times out, or shows the error.

It has nothing to do with the prompting or set up, since it's functioning fine, the next moment it does not.

Also, if the reasoning_effort cannot be set, it means this is a broken product and should be advertised accordingly.

Now, I am not alleged to a ticket and dependent on the forum. So thanks again, but it is obvious the model is broken and dysfunctional. Unfortunately, if it works, it yields the best results, so I can't use other models as an alternative.

(Maybe I'll get access to gpt-5 which hopefully will be better and, most of all, more reliable.) -> I have a customer who is not happy, understandably. I'd prefer to have a functioning o3-mini to rewriting and resetting everything for a different model...
Adam Zachary 2,265 Reputation points

2025-12-02T14:51:58.5866667+00:00

Hi Andreas,

Totally understand your frustration, especially if you are locked into an EU region for GDPR reasons. My answer was just describing what fixed a very specific pattern I was seeing, not suggesting that the service is stable everywhere.

From what I have seen so far, o3-mini with Assistants and file_search is still quite flaky on Azure: the same code can run fine for hours and then suddenly start throwing server_error on new runs with no changes. Removing reasoning_effort stopped the constant failures for me, but it does not remove all instability, and I agree that if a feature is exposed but breaks common scenarios, it should be clearly documented as a limitation.

Given your constraints, the only practical things I can suggest are:

Keep reasoning_effort off for now and add simple retry logic around runs that fail with server_error or time out.

Capture x-ms-request-id whenever it happens and ask whoever owns the Azure subscription to open a support ticket on your behalf, since only they can escalate this beyond the forum.

Treat o3-mini as preview/experimental for now and avoid depending on it for hard production SLAs until Microsoft improves the reliability or clarifies the limitations.

I agree with you on one key point: when it works, o3-mini gives excellent results, but the reliability needs to catch up to make it a safe default.
Andreas Gnyp 10 Reputation points

2025-12-02T15:02:45.2+00:00

Hi Adam,

I'm the subscription owner, but I was redirected to the forum trying to open a ticket. To my understanding I need to have an "enterprise" subscription which would cost way too much for my undertaking. Maybe I have missed something?

Thank you again!

Andreas
Adam Zachary 2,265 Reputation points

2025-12-02T15:17:11.9833333+00:00

Being the subscription owner is sufficient from a permissions point of view, but Azure only lets you open technical support tickets if there is a paid support plan attached to the account.

You do not specifically need an “Enterprise” EA subscription, but without a Developer, Standard, or higher support plan the portal will indeed push you towards docs and the community forums instead of letting you file a technical case. Microsoft Azure

Billing and subscription issues are covered for free, but what you are hitting with o3-mini is considered a technical incident, so it falls under those paid plans. In smaller projects that cost can absolutely be hard to justify, which is why many of us end up relying on Q&A and GitHub issues to surface these reliability problems so the product team can see the pattern.

Share via

o3-mini Server Error: "Sorry, something went wrong."

1 answer

Your answer