Configuring Azure API Management to Inject GPT-5 Mini Model Name Using Terraform

Steps

1. Create the API Operation
   Define the API operation that forwards requests to the Azure OpenAI Responses API endpoint.

2. resource "azurerm_api_management_api_operation" "chat_completion_app_bpp" {
     operation_id        = "ChatCompletions_Create"
     api_name            = azurerm_api_management_api.apim-api-app-bpp.name
     resource_group_name = module.ai_foundations_platform.resource_group_name
     api_management_name = module.ai_foundations_platform.apim_name
   
     display_name = "Creates a completion for the chat message"
     method       = "POST"
   
     url_template = "/openai/responses?api-version=2025-04-01-preview"
   
     description = "Creates a completion for the chat message."
   }

3. Configure an APIM Operation Policy
   Use an API operation policy to modify the inbound request and inject the required model deployment name.

4. resource "azurerm_api_management_api_operation_policy" "chat_completion_policy" {
     api_name            = azurerm_api_management_api.apim-api-app-bpp.name
     api_management_name = module.ai_foundations_platform.apim_name
     resource_group_name = module.ai_foundations_platform.resource_group_name
     operation_id        = azurerm_api_management_api_operation.chat_completion_app_bpp.operation_id
   
     xml_content = <<XML
   <policies>
     <inbound>
       <base />
   
       <!-- Example policy logic to inject model deployment name -->
       <set-body template="liquid">
       {
         "model": "gpt-5-mini-deployment",
         "input": {{body.input}}
       }
       </set-body>
   
     </inbound>
   
     <backend>
       <base />
     </backend>
   
     <outbound>
       <base />
     </outbound>
   
     <on-error>
       <base />
     </on-error>
   </policies>
   XML
   }

5. Validate the Request
   After applying the Terraform configuration:
   Send a POST request through the APIM endpoint.
   Verify that the request body forwarded to Azure OpenAI contains:

6. {
     "model": "gpt-5-mini-deployment",
     "input": "Hello"
   }

7. 
   Confirm that Azure OpenAI successfully processes the request using the specified deployment.