Azure AI Foundry model deployment fails with 715-123420 for both Bicep and Portal

Hello @Praveen Kumar Pudi

Welcome to Microsoft Q&A .Thank you for reaching out to us.

 In addition to the inputs provided by Jerald Felix , please check if the following help

The consistent failure pattern across models, regions, and deployment methods provides strong insight into where the issue is occurring.

Model deployment failures are consistently observed across:

• Multiple models
• Multiple regions
• Multiple deployment methods

At the same time, resource creation (account, project, identity) succeeds without issues.

This pattern indicates that requests are successfully reaching the service but are failing during backend validation at the deployment stage, specifically for Microsoft.CognitiveServices/accounts/deployments.

Based on this behavior, the issue is most likely related to:

• Subscription-level entitlement not enabled for the requested models, or
• Quota not allocated for the subscription in the selected regions, or
• Service-side validation blocking deployment provisioning

In such scenarios, deployments may fail with generic errors like 715-123420, even when configuration and templates are correct.

Please check if the following steps help -

1. Confirming model availability - please run the command below to verify which models are enabled az cognitiveservices account list-models -n <accountName> -g <resourceGroup> If intended models (for example, gpt-5.4, text-embedding-3-*) are not listed, deployment will fail due to missing entitlement
2. Validating quota allocation From the Foundry resource - Quota, verify:
   • Tokens per minute (TPM)
   • Requests per minute (RPM)
   • If quota is zero or not assigned, deployments cannot proceed and may return generic errors
3. Validating region and model alignment
   1. Ensure selected models are supported in the chosen regions
   2. If required, test in regions with broader availability (for example: Sweden Central, South Central US, West US 3)
4. Validating resource configuration
   1. Confirm resource kind is AIServices
   2. Ensure deployment uses a supported API version (2024‑10‑01 or later)
5. Checking policy and deployment restrictions
   1. Review Azure Policy / deny assignments
   2. Confirm there are no restrictions on Microsoft.CognitiveServices/accounts/deployments

The following references might be helpful , please check them out

• Deploy models using Azure CLI and Bicep - Microsoft Foundry | Microsoft Learn
• Feature availability across cloud regions - Microsoft Foundry | Microsoft Learn
• How to troubleshoot your deployments and monitors in Microsoft Foundry portal (classic) - Microsoft Foundry (classic) portal | Microsoft Learn
• Azure OpenAI in Microsoft Foundry Models Quotas and Limits - Microsoft Foundry | Microsoft Learn

Thank you

Hello Praveen Kumar Pudi,

Greetings!

Thanks for raising this question in Q&A forum.

You've done an excellent job of isolating this — the fact that error 715-123420 occurs across both Bicep/ARM and the Azure Portal, across multiple models, and across two different regions (West Europe and East US 2), strongly points to a service-side issue at the Microsoft.CognitiveServices/accounts/deployments resource level. This is not something caused by your templates or configuration.

That said, here are some things worth verifying and trying before escalating to support:

Step 1: Confirm the Correct Resource Kind for Azure AI Foundry Deployments Azure AI Foundry (the new unified experience) uses a different resource kind than classic Azure OpenAI. Make sure your Bicep is targeting kind: "AIServices" and not kind: "OpenAI". Using the wrong kind can silently block deployments at the provisioning stage. Here's a quick reference snippet:

resource foundryAccount 'Microsoft.CognitiveServices/accounts@2024-10-01' = {
  name: accountName
  location: location
  kind: 'AIServices'
  sku: {
    name: 'S0'
  }
  properties: {
    customSubDomainName: accountName
  }
}

Step 2: Verify the API Version for the Deployment Resource Older API versions may not support the newer Foundry-style model deployments. Make sure you are using at least 2024-10-01 or later for Microsoft.CognitiveServices/accounts/deployments. Using an older API version like 2023-05-01 can cause validation failures with newer models like gpt-5.4.

Step 3: Check That the Foundry Account and Deployment Are in the Same Resource Group and Subscription Even though account and project creation succeeds, deployment provisioning can fail silently if there's a policy restriction (like Azure Policy denying specific child resource types). Check your subscription's Policy compliance blade for any deny assignments targeting Microsoft.CognitiveServices/accounts/deployments.

Step 4: Validate Quota at the Model + Region Level Some of the models you attempted (like gpt-5.4 and gpt-5.4-nano) may have very limited or zero quota in West Europe and East US 2 at this time. Go to Azure Portal → Subscriptions → Usage + Quotas → filter for Azure OpenAI/AI Services and check per-model TPM availability for each region.

Step 5: Try with a Minimal Deployment via REST API To completely rule out any Bicep/portal layer issues, try a raw REST call using a tool like Postman or curl to directly call the deployments endpoint. This helps confirm whether the error is truly service-side.

Step 6: Open an Azure Support Ticket (Recommended) Given the consistent failure across all models, both regions, and both deployment methods, this is very likely a backend service issue tied to your subscription or tenant. When raising the ticket, please include:

• Subscription ID
• Foundry Account resource ID
• Timestamps of failed deployment attempts
• Error code: 715-123420
• All models and regions attempted

This will allow the Azure support team to check backend provisioning logs and identify the root cause quickly.

If this answer helps you kindly accept the answer which will help others who have similar questions.

Best Regards,

Jerald Felix.