o3-mini Model Error

Question

o3-mini Model Error

Rizki Susanto 0

Trying to test a request with o3-mini models but encountering the following error:

"error": 
{ 
"code": "InternalServerError", "error": 
{ 
"code": "InternalServerError", 
"message": "Gateway cannot authenticate upstream services. Please contact Microsoft for help." 
},

Other models like 4o and Deepsek R1 are working fine. The same issue occurs in the Foundry playground:

InternalServerError: Gateway cannot authenticate upstream services. Please contact Microsoft for help. | Apim-request-id: 0eb8749f-35fd-4d98-b3ee-6693999e25bf

Manas Mohanty 2,030 Reputation points Microsoft External Staff

2025-03-10T14:45:02.65+00:00

Hi Rizki Susanto

Is your Azure AI foundry resource protected with virtual network.

I checked a public OpenAI resource for 03 min in East Us 2, was able to get response in time.

Could you confirm your region too.

Thank you.
Ibis Prevedello 0 Reputation points

2025-03-11T00:34:49.57+00:00

I'm getting the same error.

How to check resource protected with virtual network?

Deployment type: Global Standard
Resource location: East US 2

As @Rizki Susanto mentioned, other models are working fine, even in playground.
Manas Mohanty 2,030 Reputation points Microsoft External Staff

2025-03-11T09:25:35.47+00:00

Hi Rizki Susanto and Ibis Prevedello

You can check for availability of private endpoint or firewall whitelisting with selected virtual network from firewall setting of your Azure OpenAI or foundry from portal.azure.com

I am able to get answer the answer slowly and bit faster while adjusting max_tokens and reasoning in a public resource.

If the resource is virtual network, we might need to do additional whitelisting of hosts in org firewall or azure firewall, or proxy

If it is originating from SDK or API usage, please make sure you are sending correct header including the API key and endpoint

If it is a backend issue, please try again later to see if it is intermittent issues

Thank you.
Rizki Susanto 0 Reputation points

2025-03-11T16:48:42.2533333+00:00

Networking firewalls have been set to allow requests from all networks as shown above. I'm using East US 2 as a region and Global Standard for deployment type.
Manas Mohanty 2,030 Reputation points Microsoft External Staff

2025-03-11T17:24:39.48+00:00

Hi Rizki Susanto

Are you using "Add your own data scenario" somehow with your 03 model.

It is not a supported featured for o3 and o1 models yet.

Thank you.
Rizki Susanto 0 Reputation points

2025-03-11T17:27:54.8133333+00:00

Hi

I didn't make any changes/settings to the deployed model, I directly deployed with the conditions that were automatically filled in by azure. I can say that this is a fresh deployment, but it doesn't work if I request on the playground or using the API.

Thanks!
kothapally Snigdha 1,945 Reputation points Microsoft External Staff

2025-03-12T01:14:19.3166667+00:00
Hi Rizki Susanto

Could you please share the below details in private message.

Azure subscription ID

Azure resource URL

Azure region - I think is EastUS2

Input prompt

Deployment Type (Pay-go / PTU / PTU-Managed

JAYA SHANKAR G S 1,450 Microsoft External Staff

Hi @Rizki Susanto ,

I have tried to reproduce the issue, but i am able to get results successfully in the same region you mentioned.

User's image

Also, got results while using sdk.

User's image

Please try it again with fresh deployment using below sdk code once and let me know the results.


import os  
import base64
from openai import AzureOpenAI  
from azure.identity import DefaultAzureCredential, get_bearer_token_provider,ClientSecretCredential

endpoint = os.getenv("ENDPOINT_URL", "<your_endpoint>")  
deployment = os.getenv("DEPLOYMENT_NAME", "o3-mini")  
subscription_key = os.getenv("AZURE_OPENAI_API_KEY", "Api_key")
client = AzureOpenAI(  
    azure_endpoint=endpoint,  
    api_key=subscription_key,  
    api_version="2024-12-01-preview",
)
    

chat_prompt = [
    {
        "role": "user",
        "content": "give me details of bank."
    }
]  
messages = chat_prompt   
completion = client.chat.completions.create(  
    model=deployment,
    messages=messages,
    max_completion_tokens=100000,
    stop=None,  
    stream=False
)
print(completion.to_json())

Thank you

Rizki Susanto 0 Reputation points

2025-03-12T17:28:39.77+00:00

After redeploying it seems to work now. However, this will create a new resource which is a bit confusing. Maybe Azure can consider simplicity regarding endpoint deployment so that it is on the same endpoint.

1 answer

Your answer

Manas Mohanty 2,030 Reputation points Microsoft External Staff

2025-03-10T14:45:02.65+00:00

Hi Rizki Susanto

Is your Azure AI foundry resource protected with virtual network.

I checked a public OpenAI resource for 03 min in East Us 2, was able to get response in time.

Could you confirm your region too.

Thank you.
Ibis Prevedello 0 Reputation points

2025-03-11T00:34:49.57+00:00

I'm getting the same error.

How to check resource protected with virtual network?

Deployment type: Global Standard
Resource location: East US 2

As @Rizki Susanto mentioned, other models are working fine, even in playground.
Manas Mohanty 2,030 Reputation points Microsoft External Staff

2025-03-11T09:25:35.47+00:00

Hi Rizki Susanto and Ibis Prevedello

You can check for availability of private endpoint or firewall whitelisting with selected virtual network from firewall setting of your Azure OpenAI or foundry from portal.azure.com

I am able to get answer the answer slowly and bit faster while adjusting max_tokens and reasoning in a public resource.

If the resource is virtual network, we might need to do additional whitelisting of hosts in org firewall or azure firewall, or proxy

If it is originating from SDK or API usage, please make sure you are sending correct header including the API key and endpoint

If it is a backend issue, please try again later to see if it is intermittent issues

Thank you.
Rizki Susanto 0 Reputation points

2025-03-11T16:48:42.2533333+00:00

Networking firewalls have been set to allow requests from all networks as shown above. I'm using East US 2 as a region and Global Standard for deployment type.
Manas Mohanty 2,030 Reputation points Microsoft External Staff

2025-03-11T17:24:39.48+00:00

Hi Rizki Susanto

Are you using "Add your own data scenario" somehow with your 03 model.

It is not a supported featured for o3 and o1 models yet.

Thank you.
Rizki Susanto 0 Reputation points

2025-03-11T17:27:54.8133333+00:00

Hi

I didn't make any changes/settings to the deployed model, I directly deployed with the conditions that were automatically filled in by azure. I can say that this is a fresh deployment, but it doesn't work if I request on the playground or using the API.

Thanks!
kothapally Snigdha 1,945 Reputation points Microsoft External Staff

2025-03-12T01:14:19.3166667+00:00

Hi Rizki Susanto

Could you please share the below details in private message.

Azure subscription ID

Azure resource URL

Azure region - I think is EastUS2

Input prompt

Deployment Type (Pay-go / PTU / PTU-Managed
JAYA SHANKAR G S 1,450 Reputation points Microsoft External Staff

2025-03-12T10:39:32.0066667+00:00

Hi @Rizki Susanto ,

I have tried to reproduce the issue, but i am able to get results successfully in the same region you mentioned.

Also, got results while using sdk.

Please try it again with fresh deployment using below sdk code once and let me know the results.

import os import base64 from openai import AzureOpenAI from azure.identity import DefaultAzureCredential, get_bearer_token_provider,ClientSecretCredential endpoint = os.getenv("ENDPOINT_URL", "<your_endpoint>") deployment = os.getenv("DEPLOYMENT_NAME", "o3-mini") subscription_key = os.getenv("AZURE_OPENAI_API_KEY", "Api_key") client = AzureOpenAI( azure_endpoint=endpoint, api_key=subscription_key, api_version="2024-12-01-preview", ) chat_prompt = [ { "role": "user", "content": "give me details of bank." } ] messages = chat_prompt completion = client.chat.completions.create( model=deployment, messages=messages, max_completion_tokens=100000, stop=None, stream=False ) print(completion.to_json())

Thank you
Rizki Susanto 0 Reputation points

2025-03-12T17:28:39.77+00:00

After redeploying it seems to work now. However, this will create a new resource which is a bit confusing. Maybe Azure can consider simplicity regarding endpoint deployment so that it is on the same endpoint.

Answer 1

JAYA SHANKAR G S 1,450 Microsoft External Staff

Hi @Rizki Susanto ,

That's good to here, now you are able to get results from 03-mini.

The reason it creates new resource while deploying is because the current ai service region is not supported with the model's deployment region.

If you observe this document, for Global standard deployment o3-mini and o1 are not available in eastus, but available in eastus2.

So, whenever you trying to deploy the model if there is no ai services available in the supported region it creates new one.

If the above solution helps, please do accept it and give feedback clicking yes.

Thank you

JAYA SHANKAR G S 1,450 Reputation points Microsoft External Staff

2025-03-14T03:49:37.82+00:00

Hi @Rizki Susanto ,

Just checking whether above solution helped you or not.

Please let me know if you have any query.

Thank you
JAYA SHANKAR G S 1,450 Reputation points Microsoft External Staff

2025-03-17T03:41:49.05+00:00

Hi @Rizki Susanto ,

Hope the given solution helpful to you, please do accept and give feedback by clicking on yes.

Thank you

Share via

o3-mini Model Error

1 answer

Your answer