Quotas and limits for model router

Model router in Microsoft Foundry has its own quota, separate from the quota for the models it routes to. You size one limit for each deployment, no matter which models answer. This article explains how much quota you get, how it's scoped, how routed requests count against it, and how to check, change, and increase it. It also covers the limit for Auto (model router without a deployment) and the other limits that apply.

  • Each model router deployment has a tokens-per-minute (TPM) limit and a requests-per-minute (RPM) limit. They apply to every request the deployment routes, whichever model answers.
  • You assign quota to deployments in units. One unit is 1,000 TPM and 1 RPM, and a new deployment gets 10 units by default.
  • Global Standard model router quota is one pool per subscription, shared across all regions. Data Zone Standard quota is one pool per subscription in each data zone.
  • Auto (preview) has its own limit, which depends on your resource's region and your subscription's model router quota.
  • You don't need quota for the underlying models, except for your own deployments that serve routed requests (preview) and your Claude deployments.

Important

Features or capabilities marked (preview) in this article are in public preview and are subject to the Supplemental Terms of Use for Microsoft Azure Previews. Previews are provided without a service-level agreement, and Microsoft doesn't recommend them for production workloads. Certain features might not be supported or might have constrained capabilities.

How model router quota works

Model router quota has its own names, one for each deployment type. You assign it to model router deployments in units of capacity.

Deployment type Quota name One unit of capacity Scope
Global Standard OpenAI.GlobalStandard.ModelRouter 1,000 TPM and 1 RPM One pool per subscription, shared across all regions.
Data Zone Standard OpenAI.DataZoneStandard.ModelRouter 1,000 TPM and 1 RPM One pool per subscription in each data zone, for example US or EU.
  • When you create a model router deployment, you set its capacity in units (sku.capacity). The default is 10 units (10,000 TPM and 10 RPM). That capacity comes out of the matching pool: your subscription's Global Standard pool, or the Data Zone Standard pool for the deployment's data zone.
  • Because the Global Standard pool is shared across regions, Global Standard model router deployments in every region draw from the same quota.
  • The deployment's limits apply to every request it routes, whichever model answers. You don't need quota for the underlying models, except for your own deployments that serve requests and your Claude deployments.
  • Auto doesn't use a deployment, so its limit works differently. See Quota for Auto.

Quota by usage tier

Your subscription's model router quota depends on its usage tier. For how tiers work, see Quota tiers. As of November 4, 2026, the model router limits are:

Tier Global Standard RPM Global Standard TPM Data Zone Standard RPM Data Zone Standard TPM
Tier 1 1,000 1,000,000 300 300,000
Tier 2 2,000 2,000,000 670 670,000
Tier 3 4,000 4,000,000 1,000 1,000,000
Tier 4 7,000 7,000,000 2,000 2,000,000
Tier 5 10,000 10,000,000 3,000 3,000,000
Tier 6 15,000 15,000,000 4,000 4,000,000

Subscriptions in the Free tier (Tier 0) have no model router quota.

Check your quota

  1. In the Foundry portal, select Manage in the upper-right navigation.
  2. Select Quota, and then select the Token per minute tab.
  3. Find the model router entries. They show how much quota is assigned to deployments and how much is still available.

For the full walkthrough, see View and request quotas in Foundry portal.

Change a deployment's limit

To give a model router deployment more or less capacity, change its quota allocation on the Quota page. Select the deployment, and then edit its allocation under Affiliated deployments using shared quota. For the steps, see View and request quotas in Foundry portal.

To set capacity with the REST API or Azure CLI, set sku.capacity in the deployment request, in units of 1,000 TPM. To change an existing deployment, send its PUT request again with the new value. For examples, see Set capacity.

Increase your quota

To get more model router quota than your tier allows, submit a quota request. Requests are reviewed manually.

  1. Open the quota request form.
  2. For the quota type, select Model Deployment.
  3. For the model type, select Azure OpenAI.
  4. Choose your deployment type: Global Standard or Data Zone Standard.
  5. For the model, select ModelRouter.
  6. Complete the rest of the form, and then submit it.

After the increase is approved, assign the new quota to your deployments. A quota increase also raises Auto's limit. See Quota for Auto.

Quota for Auto

Auto (preview) is model router without a deployment: you set model to auto in the request. It has no sku.capacity to set. Its limit depends on your resource's region:

Your resource's region Auto's limit How to get more
West US 3 Your subscription's Global Standard model router quota that isn't assigned to deployments. Reduce or delete unused model router deployments to free quota, or request a quota increase.
Every other model router region 1% of your subscription's model router quota. At Tier 1 (1,000,000 TPM), that's 10,000 TPM and 10 RPM. It grows with your quota tier and with quota increases. It isn't shown in the portal. Request a quota increase, or deploy model router for dedicated capacity.

In West US 3, Global Standard model router deployments in any region reduce Auto's limit, because they draw from the same subscription-wide pool.

To see Auto's current limit, check the x-ratelimit-limit-tokens and x-ratelimit-limit-requests headers on any Auto response. The x-ratelimit-remaining-tokens and x-ratelimit-remaining-requests headers show what's left in the current window.

If Auto requests fail with the error Insufficient quota available for instant inference, all of your subscription's Global Standard model router quota is assigned to deployments. To fix it, reduce or delete unused model router deployments, or request more quota.

Auto always processes data as Global Standard. For production traffic, higher limits, a custom content filter, or Data Zone Standard, deploy your own model router. To compare the two, see Choose Auto or your own deployment.

Quota when your deployments serve requests

Your own deployments serve routed requests first (preview). After model router picks a model, it serves the request from your own deployment of that model when one exists in the same Foundry resource with the exact model name and version: first a provisioned throughput (PTU) deployment (Global Provisioned, Data Zone Provisioned, or Regional Provisioned) for OpenAI models, and then a pay-as-you-go deployment (Standard, Global Standard, or Data Zone Standard). When your deployment serves a request:

  • On a provisioned deployment, the request uses that deployment's PTUs. You pay only the model router routing fee, because you already pay for the PTUs.
  • On a pay-as-you-go deployment, the request counts against that deployment's quota and rate limits. You pay that deployment's usage, plus the routing fee.
  • The request also counts against your model router deployment's limits, which apply to every request it routes.
  • If your deployment returns a retryable error, such as HTTP 429, model router moves down the capacity order for the same model: from your provisioned deployment to your pay-as-you-go deployment, and then to router-managed capacity. If those also fail, it fails over to the next-ranked model.

Responses don't show which capacity served a request. To make a deployment eligible, see Use your own deployments for routed requests.

When you reach the limit

Requests that exceed a deployment's limits fail with HTTP status code 429. Every response includes headers that show the deployment's limits and what remains in the current window:

Header Meaning
x-ratelimit-limit-tokens The deployment's TPM limit.
x-ratelimit-remaining-tokens Tokens left in the current window.
x-ratelimit-limit-requests The deployment's RPM limit.
x-ratelimit-remaining-requests Requests left in the current window.
x-ratelimit-renewalperiod-tokens, x-ratelimit-renewalperiod-requests The window length, in seconds.

To reduce 429 errors:

Other limits

Limit Details What to do
Context window The smallest context window among the models in your pool. A request with a larger context succeeds only if it's routed to a model that supports it. Use a model subset of models that support the context length you need.
Maximum output tokens Depends on the model that serves the request. Set an output limit that every model in your pool supports.
Input types Text and images. Model router bases its routing decision on the text only. Audio input isn't supported. Use a direct model deployment for audio.
Model parameters When the selected model doesn't support a parameter, such as temperature on a reasoning model, model router drops it. For details, see How model router handles parameters.
Claude models Claude deployments have their own quotas and rate limits. See Claude models quotas and rate limits.