Unable to access model quotas in Azure AI Foundry despite $5,000 in credits

Quota and credits are separate systems. Azure credits (for example from offers or sponsorships) cover billing, but they do not automatically grant quota or capacity for specific Azure OpenAI / Foundry models. Modern models (gpt‑4o, gpt‑4.1, gpt‑5, etc.) are controlled by:

• Per‑subscription, per‑region, per‑model TPM/RPM quota
• Regional capacity and model availability
• Sometimes gating/allow‑listing for certain models

Because of this, it is possible to have large unused credits but still see:

• Only legacy models (like GPT‑3.5) with quota
• Newer models locked or showing 0 quota in most or all regions

Below is how to diagnose and then request or unlock quota.

1. Confirm model availability and regional scope

• Quota is defined per subscription, per region, per model or deployment type (Standard, GlobalStandard, DataZoneStandard, etc.).
• If a model is not available in a region, it will not have quota there at all.
• Switching regions is useful only where the model is actually offered; otherwise it will remain locked.

See: regional quota allocation and model availability notes in the quotas and limits documentation.

2. Check current quota and capacity correctly Use the Azure AI Foundry portal first:

• Go to Azure AI Foundry → Management → Quota.
• Verify the correct subscription and region are selected.
• For each model family (for example gpt‑4o, gpt‑4.1, gpt‑5), check:
  • Quota type (Standard, GlobalStandard, DataZoneStandard, etc.)
  • Quota allocation (used vs total)

If the Quota page appears empty or does not show allocations for models you expect:

• Ensure the subscription has the Cognitive Services Usages Reader role assigned at the subscription level. Without this, quota information can be incomplete or missing.
  • Subscriptions → Access control (IAM) → Add role assignment → Cognitive Services Usages Reader.

3. Programmatically verify quota and capacity If the portal view is confusing or seems inconsistent, use the management APIs to see what is actually allocated.

a) Quota usage (per model/deployment type)

• Use the quota/usage API (for example via az rest) to list usages and limits.
• Key fields:
  • name.value: {Provider}.{DeploymentType}.{Model} (for example OpenAI.Standard.gpt-4o).
  • currentValue: how much quota is currently consumed.
  • limit: the quota limit for that model/deployment type.

If limit is 0 for gpt‑4o/gpt‑5 families, that means no quota has been granted yet for those models in that subscription/region.

b) Model capacities (is there any capacity available to allocate?)

Use the Model Capacities API to check if your subscription has capacity for a model at all:

SUBSCRIPTION_ID="<your-subscription-id>"

az rest --method get \
  --url "https://management.azure.com/subscriptions/$SUBSCRIPTION_ID/providers/Microsoft.CognitiveServices/modelCapacities?api-version=2024-10-01&modelFormat=OpenAI&modelName=gpt-4o&modelVersion=2024-08-06"

Key fields in the response:

• location: region (for example eastus).
• properties.skuName: deployment type (Standard, GlobalStandard, DataZoneStandard, etc.).
• properties.availableCapacity: capacity units available for that model/region/type.

If availableCapacity is 0 across all regions, the subscription currently has no allocatable capacity for that model.

4. Understand why GPT‑3.5 has quota but newer models do not From the quota documentation:

• Quotas and limits are per subscription and per region, and not all models are enabled by default.
• When signing up, default quota is granted for “most of the available models,” but this can differ by subscription type and region.
• Some models (especially newer or high‑demand ones like gpt‑4o/gpt‑5) may:
  • Have 0 default quota on certain subscription types.
  • Be capacity‑constrained in some regions.
  • Require explicit access or gating approval before quota can be assigned.

This explains the situation:

• GPT‑3.5 often has default quota even on smaller or credit‑based subscriptions.
• Modern models may show as locked or 0 quota because:
  • No quota has been granted yet for that model family in that subscription/region.
  • The region is at capacity for that model.
  • The model is not enabled for that subscription (gated).

Credits do not override these constraints; they only pay for usage once quota and capacity exist.

5. Request more quota for modern models For Foundry / Azure OpenAI models, the supported process to obtain or increase quota is:
6. In Azure AI Foundry portal, go to Management → Quota.
7. For the desired model family (for example gpt‑4o, gpt‑4.1, gpt‑5) and region, select the Request Quota icon.
8. This opens the quota increase form at https://aka.ms/oai/stuquotarequest.
9. In the form, specify:
   • Subscription ID.
   • Region(s).
   • Model(s) and deployment type (Standard / GlobalStandard / DataZoneStandard).
   • Requested TPM/RPM.
   • Business justification (production workload, POC, etc.).

Notes from the documentation:

• You can continue to create deployments and assign TPM until you reach your quota limit.
• When quota is exceeded, the options are:
  • Request more quota via the form.
  • Reallocate TPM from existing deployments to free capacity.

Quota changes can take up to ~15 minutes to propagate after approval; refresh the Quota page if it does not update immediately.

6. If models remain locked after quota requests If, after submitting quota requests, the models still appear locked or with 0 quota:

• Confirm that the Request quota button is enabled. If it is disabled:
  • Ensure the account has Owner or Contributor on the subscription.
  • Some model/region combinations do not support quota increases; in those cases, another region or deployment type (for example GlobalStandard or DataZoneStandard) may be required.
• If the Quota page is empty or does not load, verify again that Cognitive Services Usages Reader is assigned at the subscription level and that the correct subscription is selected.
• If quota changes are approved but not reflected after 24 hours, open an Azure support ticket for quota/capacity investigation.

7. Practical next steps for this scenario
8. Assign Cognitive Services Usages Reader to the subscription (if not already done) and recheck the Quota page in Azure AI Foundry.
9. Use the Model Capacities API to confirm whether any capacity exists for gpt‑4o/gpt‑4.1/gpt‑5 in any region for the subscription.
10. For at least one region where the model is available, submit a quota increase request via the Request Quota button (which opens https://aka.ms/oai/stuquotarequest).
11. After approval and propagation (up to ~15 minutes), allocate TPM to new deployments of those models in that region.
12. If no capacity is available for the desired model in any region, or if the quota request is rejected, choose an alternative model/region combination that has quota and capacity, as shown in the quotas/limits and regional availability documentation.

This process is the supported way to unlock and scale quota for modern models; credits alone do not change quota or capacity.

References:

• Manage Azure OpenAI in Microsoft Foundry Models quota
• Azure OpenAI in Microsoft Foundry Models quotas and limits
• Manage Azure OpenAI in Azure AI Foundry Models quota
• Deploy Microsoft Foundry Models in the Foundry portal
• Manage and increase quotas for resources with Microsoft Foundry (Foundry projects)
• Quota shows as fully utilised for Australia East region for Standard deployment - Microsoft Q&A
• Unable to deploy resources, can't view Quotas or request increases - Microsoft Q&A