Procedure to deploy the gpt-5-mini model in a Microsoft Foundry project, test it from the portal and call it from a .NET console application authenticated with Entra ID, with no keys in the code.

Required: an Azure account with an active subscription, a browser, .NET 10 and the Azure command-line interface on the computer. Duration: about twenty minutes. Cost: billed per token; a Global Standard deployment has no fixed fee.

The portal in the screenshots is in English: command names are reported as they appear on screen.

Deploy the model

1. Create the resource group

On portal.azure.com, search for Resource groups and press + Create.

Field Value
Resource group name rg-guide-ai-01
Region (Europe) Sweden Central

Press Review + create, then Create.

Creating the resource group in the Azure portal

2. Create the project

On ai.azure.com press Create project, open Advanced options and fill in the fields.

Field Value
Project name proj-guide-01
Foundry resource proj-guide-01-resource
Region Sweden Central
Resource group rg-guide-ai-01

Leave Set up recommended resources off: when on, it also creates an Application Insights resource. Press Create; creation takes about a minute.

Create a project dialog with advanced options open

The resource group has to exist beforehand: the dialog only lets you pick one.

3. Find the model

Open Catalog, the Models tab, and type gpt-5-mini in the search box. The Available in my project filter is already on.

Model catalog filtered on gpt-5-mini

4. Open the model card

Press the gpt-5-mini card. On the right, Quick facts gives publisher, type and where it is hosted.

The gpt-5-mini model card

5. Deploy

Press Deploy and choose Default settings, that is a Global Standard deployment with the default quota.

Deploy menu with Default settings and Custom settings

Test it from the portal

6. The playground opens on the deployment

When the deployment is ready the portal opens the model playground. System instructions, tools and memory are on the left; the conversation on the right.

Playground of the newly deployed model

7. Ask a question

Type the question at the bottom and press Enter. Under the answer the portal reports the model, the time taken and the tokens used: in this run, 3 seconds and 4,186 tokens.

Model answer with time and tokens

8. Take the endpoint

Press Continue in code. The dialog gives the project endpoint and the key, and generates code in the language you pick. Set Language to C# and Authentication method to Entra ID authentication.

Field Value
Project endpoint https://proj-guide-01-resource.services.ai.azure.com/openai/v1
Deployment gpt-5-mini

Dialog with endpoint, key and generated code

Call the model from .NET

9. Create the application

dotnet new console -n foundry-demo
cd foundry-demo
dotnet add package Azure.Identity --version 1.21.0
dotnet add package OpenAI --version 2.14.0
az login

az login is needed because DefaultAzureCredential reuses the account of the command-line interface.

10. Write the code

The code generated by the portal does not compile against the OpenAI 2.14.0 package: ResponsesClient takes no model in the constructor and CreateResponseAsync wants a CreateResponseOptions. This is the version used here; it replaces Program.cs.

using Azure.Identity;
using OpenAI.Responses;
using System.ClientModel.Primitives;

#pragma warning disable OPENAI001

const string deploymentName = "gpt-5-mini";
const string endpoint = "https://proj-guide-01-resource.services.ai.azure.com/openai/v1";

BearerTokenPolicy tokenPolicy = new(
    new DefaultAzureCredential(),
    "https://ai.azure.com/.default");

ResponsesClient client = new(
    tokenPolicy,
    new ResponsesClientOptions { Endpoint = new Uri(endpoint) });

CreateResponseOptions options = new(
    deploymentName,
    [ResponseItem.CreateUserMessageItem("In one line: what is an Azure resource group?")]);

var response = await client.CreateResponseAsync(options);

Console.WriteLine(response.Value.GetOutputText());

#pragma warning disable OPENAI001 is needed because the Responses API in the package is marked experimental: without it, the build stops.

11. Run it

dotnet run

The program prints the model answer:

An Azure resource group is a logical container for related Azure resources
that allows you to deploy, manage, and apply access control, policies, and
lifecycle operations as a single unit.

If the answer is HTTP 404 (DeploymentNotFound), the deployment has just been created: wait a few minutes and try again.

Delete the resources

12. Delete the resource group

On portal.azure.com open rg-guide-ai-01. It holds two resources: the Foundry resource and the project.

The two resources inside the group

Press Delete resource group, type rg-guide-ai-01 again, press Delete, then confirm in the dialog that appears.

Delete panel with the name typed again

Once the group is gone the Foundry resource stays soft-deleted and its name cannot be reused. To free it right away:

az cognitiveservices account purge --name proj-guide-01-resource --location swedencentral --resource-group rg-guide-ai-01

Command names and screenshots taken on ai.azure.com and the Azure portal on 3 November 2026, Sweden Central region.