Automatically deploy GPT-4.1 Mini in Azure OpenAI via Bicep

The official Microsoft documentation on deploying Azure OpenAI models is generally helpful, but specific examples are often still missing, especially for new models such as gpt-4.1-mini and newer SKU types such as DataZoneStandard. Anyone who wants to provide a current model such as the gpt-4.1-mini programmatically is therefore quickly faced with unanswered questions.

Recently, some things have changed in the provision of OpenAI deployments in Azure: In particular, the SKU type DataZoneStandard is new and replaces the previous Standard or Standard_S logic for certain models. Anyone working with this new SKU needs to know how to define the capacity correctly and what to look out for when deploying.

New model availability: GPT-4.1 Mini now also available in Germany

Another important innovation: GPT-4.1 Mini is now no longer exclusively limited to data centers in France (France Central), but can also be deployed in the Germany West Central region. This is a major step forward, especially for data-sensitive projects or customers with location requirements in Germany. This makes it easier to use the model in a more data protection-friendly way while adhering to local compliance requirements.

Example deployment via Bicep

Here you can see a simple example in Bicep

 1param openAiAccountName string // z. B. 'oai-myaccountname'
 2param openAiAccountLocation string // z. B. 'germanywestcentral'
 3
 4resource openAIAccount 'Microsoft.CognitiveServices/accounts@2025-06-01' = {
 5  name: openAiAccountName
 6  location: openAiAccountLocation
 7  sku: {
 8    name: 'S0'
 9  }
10  kind: 'OpenAI'
11  properties: {
12    publicNetworkAccess: 'Enabled'
13  }
14}
15
16resource gpt4_1_mini 'Microsoft.CognitiveServices/accounts/deployments@2025-06-01' = {
17  parent: openAIAccount
18  name: '${openAiAccountName}_gpt41mini' 
19  sku: {
20    name: 'DataZoneStandard'
21    capacity: 20 
22  }
23  properties: {
24    model: {
25      format: 'OpenAI'
26      name: 'gpt-4.1-mini'
27      version: '2025-04-14' 
28    }
29    versionUpgradeOption: 'OnceNewDefaultVersionAvailable'
30    currentCapacity: 20 
31    raiPolicyName: 'Microsoft.Default'
32  }
33}

What does capacity mean with DataZoneStandard?

The capacity specification indicates how much quota unit is to be provided for the respective model. In contrast to the classic SKUs (Standard etc.), where capacities were often implicit or related to tokens, capacity here means an explicit allocation of computing capacity within the so-called DataZone. This value influences how many parallel requests and how high throughputs your deployment can process.

Important: capacity is not synonymous with a number of instances or maximum tokens - it is a model-dependent metric that is controlled and documented by Microsoft (although unfortunately not always transparently).

Conclusion

If you want to use current models such as gpt-4.1-mini in Azure OpenAI, there is no way around DataZoneStandard. It is important that you define the capacity correctly and know which model supports which version. The bicep snippet shown above is a ready-to-use template that you can use to set up a deployment in just a few minutes.

Especially in times when Microsoft releases many model versions quickly, an automated setup via Bicep is particularly worthwhile - because you can automatically update to new model versions with versionUpgradeOption: OnceNewDefaultVersionAvailable``, for example.