Region support for Azure Translator in Foundry Tools

Compare standard and LLM-based translation

Text Translation supports standard neural machine translation (NMT) and LLM-based translation. The model technology determines which resource and regional configuration applies.

The global Text Translation endpoint is:

https://api.cognitive.microsofttranslator.com

Standard NMT endpoints

Neural machine translation (NMT) supports global, geography, and resource-specific endpoints. The endpoint you use determines where Translator processes the request.

To process text translation requests in Switzerland, create the Translator resource in Switzerland North or Switzerland West. Then use the resource-specific endpoint:

https://<resource-name>.cognitiveservices.azure.com/translator/text/v3.0

When you use a regional Translator resource or a multi-service resource, include the Ocp-Apim-Subscription-Region header. For request examples and requirements by resource type, see Authentication and authorization.

LLM-based processing in Foundry

For LLM-based text translation, the model deployment type determines where the model processes translation data:

• Global deployments can process data in any Azure region where the model is deployed.
• Data zone deployments process data within the selected data zone.
• Regional deployments process data in the deployment region.

These deployment types also apply to adaptive custom translation. They don't change standard NMT endpoint routing.

For resource requirements, see Create and configure Azure resources for Translator. For processing boundaries and model availability, see Deployment types for Microsoft Foundry Models.

Plan where documents are processed

Current Document Translation APIs use NMT. The LLM-based deployment considerations in the Text Translation tab don't apply to Document Translation.

Document Translation uses the resource-specific custom endpoint:

https://<resource-name>.cognitiveservices.azure.com

The region where you create the Translator resource determines where document content is processed and temporarily stored during translation.

If you use managed identity to authorize Document Translation access to Azure Blob Storage, create the Translator resource in a geographic Azure region instead of Global. For setup requirements, see Managed identities for Document Translation.

For asynchronous batch translation, the locations of your source and target storage accounts also affect your data residency architecture. Translator temporarily stores customer data while it processes a document and deletes the data after processing. For more information, see Data, privacy, and security for Azure Translator.

Choose where to train and publish models

Custom Translator supports training and publishing custom NMT models. After you publish a model, use it with Text Translation or Document Translation. The regions in this section apply to custom model training and deployment, not to a separate inference service.

• Use the Custom Translator portal to train, test, and publish custom NMT models.

• For LLM-based customization, adaptive custom translation is available in Microsoft Foundry. It adapts translation output at runtime by using reference translations or an adaptive dataset instead of training and deploying a custom NMT model.

You can publish a trained model to one or more supported regions. Custom Translator isn't currently available in Switzerland.

The following regions and billing region codes are supported when you create a workspace with the Custom Translator API:

Virtual network scenarios require a regional Translator resource. A global resource isn't supported. Network restrictions can also prevent access to the Custom Translator portal. For configuration details, see Enable Custom Translator through Azure Virtual Network.