Azure AI Foundry: Insufficient quota error across ALL Tier 1 models despite valid Tier 1 status

Tier 1 does not by itself prove that deployable TPM has been provisioned for every model; verify the subscription’s control-plane tier and the Quota page’s available TPM for each exact model/deployment-type/scope combination, then escalate if Tier 1 is confirmed but those entries remain zero or absent.

1. Verify the subscription’s actual quota-tier assignment

Run the documented control-plane query against the same subscription selected in Foundry:

SUBSCRIPTION_ID="<subscription-id>"

curl -X GET \
  "https://management.azure.com/subscriptions/[Subscription ID]/providers/Microsoft.CognitiveServices/quotaTiers?api-version=2025-10-01-preview" \
  -H "Authorization: Bearer $(az account get-access-token \
    --resource https://management.azure.com \
    --query accessToken -o tsv)" \
  -H "Content-Type: application/json"

The expected response should contain:

{
  "properties": {
    "currentTierName": "Tier 1",
    "tierUpgradePolicy": "OnceUpgradeIsAvailable"
  }
}

This API is the documented way to check the subscription’s current quota tier. Tier eligibility is evaluated at subscription level and can also be affected by the Azure offer; some models or deployment types have offer-specific limits.^1^

If the response does not show Tier 1, the table being viewed is a tier-reference table rather than the subscription’s effective assignment. A project-level refresh cannot change that subscription-level state.

2. Check deployable quota, not the usage-tier table

In Foundry (new):

1. Select the affected project and confirm the correct subscription.
2. Go to Manage → Quota → Token per minute.
3. Find one affected model, such as gpt-6-luna.
4. Inspect each entry separately:
   • Model and version
   • GlobalStandard or DataZoneStandard
   • Scope
   • Limit, allocated TPM, and available TPM
   • Affiliated deployments using shared quota
5. Repeat for one older model such as o3-mini to establish that the issue spans separate quota pools.

The Tier 1 table states potential tier limits—for example, gpt-6-luna Global Standard at 250 RPM/250,000 TPM—but deployment requires actual available quota for that model and deployment type.^1^ Catalog visibility does not establish deployable quota.^2^

Also distinguish this from usage tiers: usage tiers describe throughput levels associated with predictable latency; they are not a blanket entitlement that automatically allocates deployable TPM to every listed model.

3. Interpret the result

• Limit is nonzero, but available TPM is zero: Open the model details and inspect affiliated deployments. Global Standard deployments of the same model/version can share one subscription-wide pool across regions; Data Zone Standard can share one pool within the data zone.^3^ Reallocate unused TPM using the pencil action if appropriate.
• Limit is 0/0 or the model entry is absent: Submit Request quota for that exact model and deployment type. Owner or Contributor is required to request quota; viewing allocations requires Cognitive Services Usages Reader at subscription scope.^4^
• Limit and available TPM are both nonzero, but every deployment still fails: This is the strongest indication of a quota-state/provisioning inconsistency rather than an allocation-size problem. Do not continue testing more models or reducing sliders.

Global Standard remains the preferred test unless data-residency requirements require Data Zone Standard; it has the broadest regional coverage and receives new models first.^5^

4. Escalate a confirmed mismatch

Quota changes can take up to 15 minutes to propagate. If an approved or nonzero allocation still does not work after 24 hours, contact Azure Support.^4^

Include:

• Subscription ID and Azure offer type
• Control-plane response showing currentTierName
• Foundry Quota screenshots/exports for two affected models
• Exact model version, deployment type, scope, limit, allocated TPM, and available TPM
• Project/resource region and data zone
• UTC timestamps and deployment correlation/request IDs
• The complete deployment error
• Confirmation that no affiliated deployments consume the displayed quota

There is no documented customer operation to force a quota synchronization. Once Tier 1 and nonzero available TPM are shown but deployment validation still rejects multiple independent models, backend investigation through Azure Support is the appropriate path.

References

1. Azure OpenAI in Microsoft Foundry Models quotas and limits
2. 0/0 TPM for all GPT-5.6 / GPT-6 models on a new startup subscription: how to get baseline quota? - Microsoft Q&A
3. Microsoft Foundry Models quotas and limits
4. Manage and increase quotas for resources - Microsoft Foundry
5. Deployment types for Microsoft Foundry Models