Merge pull request #5669 from mrbullwinkle/mrb_06_23_2025_quota

[Azure OpenAI] quota updates

Author: prmerger-automator[bot]

articles/ai-services/openai/how-to/quota.md

@@ -16,11 +16,11 @@ Quota provides the flexibility to actively manage the allocation of rate limits
 ## Prerequisites
 
 > [!IMPORTANT]
-> For any task that requires viewing available quota we recommend using the **Cognitive Services Usages Reader** role. This role provides the minimal access necessary to view quota usage across an Azure subscription. To learn more about this role and the other roles you will need to access Azure OpenAI, consult our [Azure role-based access control guide](./role-based-access-control.md). 
+> For any task that requires viewing available quota we recommend using the **Cognitive Services Usages Reader** role. This role provides the minimal access necessary to view quota usage across an Azure subscription. To learn more about this role and the other roles you'll need to access Azure OpenAI, consult our [Azure role-based access control guide](./role-based-access-control.md). 
 >
-> This role can be found in the Azure portal under **Subscriptions** > **Access control (IAM)** > **Add role assignment** > search for **Cognitive Services Usages Reader**. This role **must be applied at the subscription level**, it does not exist at the resource level.
+> This role can be found in the Azure portal under **Subscriptions** > **Access control (IAM)** > **Add role assignment** > search for **Cognitive Services Usages Reader**. This role **must be applied at the subscription level**, it doesn't exist at the resource level.
 >
-> If you do not wish to use this role, the subscription **Reader** role will provide equivalent access, but it will also grant read access beyond the scope of what is needed for viewing quota and model deployment.
+> If you don't wish to use this role, the subscription **Reader** role will provide equivalent access, but it will also grant read access beyond the scope of what is needed for viewing quota and model deployment.
 
 ## Introduction to quota
 
@@ -31,7 +31,20 @@ Azure OpenAI's quota feature enables assignment of rate limits to your deploymen
 
 When a deployment is created, the assigned TPM will directly map to the tokens-per-minute rate limit enforced on its inferencing requests. A **Requests-Per-Minute (RPM)** rate limit will also be enforced whose value is set proportionally to the TPM assignment using the following ratio:
 
-6 RPM per 1000 TPM. (This ratio can vary by model for more information, see [quota, and limits](../quotas-limits.md#o-series-rate-limits).)
+> [!IMPORTANT]
+> The ratio of Requests Per Minute (RPM) to Tokens Per Minute (TPM) for quota can vary by model. When you deploy a model programmatically or [request a quota increase](https://aka.ms/oai/stuquotarequest) you don't have granular control over TPM and RPM as independent values. Quota is allocated in terms of units of capacity which have corresponding amounts of RPM & TPM:
+>
+> | Model                  | Capacity   | Requests Per Minute (RPM)  | Tokens Per Minute (TPM) |
+> |------------------------|:----------:|:--------------------------:|:-----------------------:|
+> | **Older chat models:** | 1 Unit     | 6 RPM                      | 1,000 TPM               |
+> | **o1 & o1-preview:**   | 1 Unit     | 1 RPM                      | 6,000 TPM               |
+> | **o3**                 | 1 Unit     | 1 RPM                      | 1,000 TPM               |
+> | **o4-mini**            | 1 Unit     | 1 RPM                      | 1,000 TPM               |
+> | **o3-mini:**           | 1 Unit     | 1 RPM                      | 10,000 TPM              |
+> | **o1-mini:**           | 1 Unit     | 1 RPM                      | 10,000 TPM              |
+> | **o3-pro:**            | 1 Unit     | 1 RPM                      | 10,000 TPM              |
+>
+> This is particularly important for programmatic model deployment as changes in RPM/TPM ratio can result in accidental  misallocation of quota. For more information, see [quota, and limits](../quotas-limits.md#o-series-rate-limits).
 
 The flexibility to distribute TPM globally within a subscription and region has allowed Azure OpenAI to loosen other restrictions:
 
@@ -62,7 +75,7 @@ Different model deployments, also called model classes have unique max TPM value
 All other model classes have a common max TPM value.
 
 > [!NOTE]
-> Quota Tokens-Per-Minute (TPM) allocation is not related to the max input token limit of a model. Model input token limits are defined in the [models table](../concepts/models.md) and are not impacted by changes made to TPM.  
+> Quota Tokens-Per-Minute (TPM) allocation isn't related to the max input token limit of a model. Model input token limits are defined in the [models table](../concepts/models.md) and aren't impacted by changes made to TPM.  
 
 ## View and request quota
 
@@ -92,7 +105,7 @@ As each request is received, Azure OpenAI computes an estimated max processed-to
 As requests come into the deployment endpoint, the estimated max-processed-token count is added to a running token count of all requests that is reset each minute. If at any time during that minute, the TPM rate limit value is reached, then further requests will receive a 429 response code until the counter resets.
 
 > [!IMPORTANT]
-> The token count used in the rate limit calculation is an estimate based in part on the character count of the API request. The rate limit token estimate is not the same as the token calculation that is used for billing/determining that a request is below a model's input token limit. Due to the approximate nature of the rate limit token calculation, it is expected behavior that a rate limit can be triggered prior to what might be expected in comparison to an exact token count measurement for each request.  
+> The token count used in the rate limit calculation is an estimate based in part on the character count of the API request. The rate limit token estimate isn't the same as the token calculation that is used for billing/determining that a request is below a model's input token limit. Due to the approximate nature of the rate limit token calculation, it's expected behavior that a rate limit can be triggered prior to what might be expected in comparison to an exact token count measurement for each request.  
 
 RPM rate limits are based on the number of requests received over time. The rate limit expects that requests be evenly distributed over a one-minute period. If this average flow isn't maintained, then requests might receive a 429 response even though the limit isn't met when measured over the course of a minute. To implement this behavior, Azure OpenAI evaluates the rate of incoming requests over a small period of time, typically 1 or 10 seconds. If the number of requests received during that time exceeds what would be expected at the set RPM limit, then new requests will receive a 429 response code until the next evaluation period. For example, if Azure OpenAI is monitoring request rate on 1-second intervals, then rate limiting will occur for a 600-RPM deployment if more than 10 requests are received during each 1-second period (600 requests per minute = 10 requests per second).
 
@@ -108,7 +121,7 @@ To minimize issues related to rate limits, it's a good idea to use the following
 
 ## Automate deployment
 
-This section contains brief example templates to help get you started programmatically creating deployments that use quota to set TPM rate limits. With the introduction of quota you must use API version `2023-05-01` for resource management related activities. This API version is only for managing your resources, and does not impact the API version used for inferencing calls like completions, chat completions, embedding, image generation, etc.
+This section contains brief example templates to help get you started programmatically creating deployments that use quota to set TPM rate limits. With the introduction of quota you must use API version `2023-05-01` for resource management related activities. This API version is only for managing your resources, and doesn't impact the API version used for inferencing calls like completions, chat completions, embedding, image generation, etc.
 
 # [REST](#tab/rest)
 
@@ -139,7 +152,7 @@ This is only a subset of the available request body parameters. For the full lis
 |Parameter|Type| Description |
 |--|--|--|
 |sku | Sku | The resource model definition representing SKU.|
-|capacity|integer|This represents the amount of [quota](../how-to/quota.md) you are assigning to this deployment. A value of 1 equals 1,000 Tokens per Minute (TPM). A value of 10 equals 10k Tokens per Minute (TPM).|
+|capacity|integer|This represents the amount of [quota](../how-to/quota.md) you're assigning to this deployment. A value of 1 equals 1,000 Tokens per Minute (TPM). A value of 10 equals 10k Tokens per Minute (TPM).|
 
 #### Example request
 
@@ -186,7 +199,7 @@ curl -X GET https://management.azure.com/subscriptions/00000000-0000-0000-0000-0
 
 Install the [Azure CLI](/cli/azure/install-azure-cli). Quota requires `Azure CLI version 2.51.0`. If you already have Azure CLI installed locally run `az upgrade` to update to the latest version.
 
-To check which version of Azure CLI you are running use `az version`. Azure Cloud Shell is currently still running 2.50.0 so in the interim local installation of Azure CLI is required to take advantage of the latest Azure OpenAI features.
+To check which version of Azure CLI you're running use `az version`. Azure Cloud Shell is currently still running 2.50.0 so in the interim local installation of Azure CLI is required to take advantage of the latest Azure OpenAI features.
 
 ### Deployment
 
@@ -239,7 +252,7 @@ For more information, see the [Azure CLI reference documentation](/cli/azure/cog
 
 Install the latest version of the [Az PowerShell module](/powershell/azure/install-azure-powershell). If you already have the Az PowerShell module installed locally, run `Update-Module -Name Az` to update to the latest version.
 
-To check which version of the Az PowerShell module you are running, use `Get-InstalledModule -Name Az`. Azure Cloud Shell is currently running a version of Azure PowerShell that can take advantage of the latest Azure OpenAI features.
+To check which version of the Az PowerShell module you're running, use `Get-InstalledModule -Name Az`. Azure Cloud Shell is currently running a version of Azure PowerShell that can take advantage of the latest Azure OpenAI features.
 
 ### Deployment
 

articles/ai-services/openai/quotas-limits.md

@@ -99,30 +99,29 @@ The following sections provide you with a quick guide to the default quotas and
 | `model-router` (2025-05-19) | Enterprise Tier | 10 M | 10 K |
 | `model-router` (2025-05-19) | Default         | 1 M | 1 K |
 
-
 ## computer-use-preview global standard rate limits
 
 | Model|Tier| Quota Limit in tokens per minute (TPM) | Requests per minute |
 |---|---|:---:|:---:|
 | `computer-use-preview`| Enterprise Tier | 30 M | 300 K |
 | `computer-use-preview`| Default         | 450 K | 4.5 K |
 
-
 ## o-series rate limits
 
 > [!IMPORTANT]
-> The ratio of RPM/TPM for quota with o1-series models works differently than older chat completions models:
->
-> - **Older chat models:** 1 unit of capacity = 6 RPM and 1,000 TPM.
-> - **o1 & o1-preview:** 1 unit of capacity = 1 RPM and 6,000 TPM.
-> - **o3** 1 unit of capacity = 1 RPM per 1,000 TPM
-> - **o4-mini** 1 unit of capacity = 1 RPM per 1,000 TPM
-> - **o3-mini:** 1 unit of capacity = 1 RPM per 10,000 TPM.
-> - **o1-mini:** 1 unit of capacity = 1 RPM per 10,000 TPM.
+> The ratio of Requests Per Minute (RPM) to Tokens Per Minute (TPM) for quota can vary by model. When you deploy a model programmatically or [request a quota increase](https://aka.ms/oai/stuquotarequest) you don't have granular control over TPM and RPM as independent values. Quota is allocated in terms of units of capacity which have corresponding amounts of RPM & TPM:
 >
-> This is particularly important for programmatic model deployment as this change in RPM/TPM ratio can result in accidental under allocation of quota if one is still assuming the 1:1000 ratio followed by older chat completion models.
+> | Model                  | Capacity   | Requests Per Minute (RPM)  | Tokens Per Minute (TPM) |
+> |------------------------|:----------:|:--------------------------:|:-----------------------:|
+> | **Older chat models:** | 1 Unit     | 6 RPM                      | 1,000 TPM               |
+> | **o1 & o1-preview:**   | 1 Unit     | 1 RPM                      | 6,000 TPM               |
+> | **o3**                 | 1 Unit     | 1 RPM                      | 1,000 TPM               |
+> | **o4-mini**            | 1 Unit     | 1 RPM                      | 1,000 TPM               |
+> | **o3-mini:**           | 1 Unit     | 1 RPM                      | 10,000 TPM              |
+> | **o1-mini:**           | 1 Unit     | 1 RPM                      | 10,000 TPM              |
+> | **o3-pro:**            | 1 Unit     | 1 RPM                      | 10,000 TPM              |
 >
-> There's a known issue with the [quota/usages API](/rest/api/aiservices/accountmanagement/usages/list?view=rest-aiservices-accountmanagement-2024-06-01-preview&tabs=HTTP&preserve-view=true) where it assumes the old ratio applies to the new o1-series models. The API returns the correct base capacity number, but doesn't apply the correct ratio for the accurate calculation of TPM.
+> This is particularly important for programmatic model deployment as changes in RPM/TPM ratio can result in accidental  misallocation of quota.
 
 ### o-series global standard