What is Azure text translation in Microsoft Foundry?

Azure text translation is a cloud-based REST API for multilingual translation across more than 100 supported languages and dialects. You can integrate it into applications, services, and automation workflows to translate text in real time at production scale by using REST APIs or SDKs.

The latest generally available release is 2026-06-06. This version introduces a revised request and response schema, adds support for model selection between standard Neural Machine Translation (NMT) and supported Large Language Models (LLMs), and expands translation controls for scenarios that require adaptive output, tone, and gender-specific variations.

Translator is designed for high-volume multilingual workloads where consistency, scalability, and operational control matter. If you're currently on v3.0, plan migration and validation before rollout because 2026-06-06 includes breaking changes.

What's new in 2026-06-06 (GA)

API version 2026-06-06 is generally available and includes breaking changes compared to v3.0.

This release adds new translation controls and model options while updating request and response contracts. If you're migrating from v3.0, plan a full validation pass for payload shape, response parsing, and production quality checks before rollout.

Key updates in 2026-06-06 include:

• Revised request and response JSON schema. The request array uses inputs, and the response array uses value. For an end-to-end example, see REST API guide.
• Model choice (NMT or LLM). You can select standard NMT or an LLM deployment (for example, GPT-5.1) based on quality, cost, and scenario requirements. Using LLM-based translation requires a Microsoft Foundry resource. For setup guidance, see Create Translator resources.
• Adaptive custom translation. You can provide up to five reference translations or an adaptive dataset index ID to influence LLM output style and terminology. For more information, see Adaptive custom translation.
• Tone and gender controls. LLM-based translation supports tone variants (formal, informal, neutral) and gender-specific output controls.

Core operations

The 2026-06-06 API includes the following primary operations:

These operations cover the most common runtime needs for text localization pipelines: discovering supported languages, translating text at scale, and transliterating scripts for language pairs that require script conversion.

• Languages: Returns supported languages for translation operations. This request does not require authentication.

  https://api.cognitive.microsofttranslator.com/languages?api-version=2026-06-06
  

• Translate: Translates source text to one or more target languages in a single POST request.

• Transliterate: Converts script or character sets from one writing system to another.

For a complete request walkthrough, see REST API guide.

Development options

Use the following options to build and deploy multilingual text translation solutions in your applications and workflows:

Choose your integration path based on how much control and automation you need. Most production systems use direct REST API or SDK integration, while portal workflows can be useful for setup, testing, and operational troubleshooting.

Authentication

All text translation requests require authentication headers. For production deployments, choose an approach that matches your security and operations model. Resource keys are simple to start with, while Microsoft Entra ID is typically preferred for enterprise environments that require centralized identity and keyless access.

For required headers and endpoint-specific examples, see Authentication and authorization.

Endpoints and data residency

Requests are typically processed in the nearest available datacenter when you use the global endpoint. To constrain processing geography, use a regional endpoint.

Endpoint selection affects latency, geography, and regulatory posture. Use the global endpoint for broad availability and automatic routing, or use a regional endpoint when your architecture requires processing to remain within a specific geography.

Switzerland service endpoints

If your Translator resource is deployed in Switzerland North or Switzerland West, you can keep text translation processing in Switzerland by using the resource-specific custom endpoint.

This configuration is useful for workloads with stricter geographic processing requirements. Ensure your Translator resource region and request endpoint remain aligned so requests are routed within Switzerland.

Example request:

curl -X POST "https://my-swiss-n.cognitiveservices.azure.com/translator/text/v3.0/translate?to=fr" \
  -H "Ocp-Apim-Subscription-Key: ${TRANSLATOR_KEY}" \
  -H "Ocp-Apim-Subscription-Region: switzerlandnorth" \
  -H "Content-Type: application/json" \
  -d "[{\"Text\":\"Hello\"}]"

Custom Translator is not currently available in Switzerland.

LLM processing, limits, and pricing

When you deploy an LLM, your deployment configuration (global, data zone, or regional) determines where LLM translation data is processed.

For mixed NMT and LLM deployments, treat model routing, limits, and cost as part of your runtime design. Throughput and payload sizing differ between NMT and LLM paths, so baseline both latency and cost before scaling traffic.

Service limits

For broader quotas and limits, see Service limits - Translator.

Pricing

• NMT translation is billed by source-text characters. See Azure Translator pricing.
• LLM translation is billed by processed input and output tokens. See Azure OpenAI pricing.

If you support multiple traffic tiers, define routing rules that reserve LLM use for scenarios that require adaptive behavior, tone control, or richer contextual handling, and keep high-volume baseline traffic on NMT where appropriate.

Troubleshooting

If a request fails, start with the HTTP status and the service error code.

In production systems, the fastest way to isolate issues is to log request IDs, endpoint, region, API version, and key headers with each failed call. This context helps distinguish authentication, quota, payload, and feature-availability failures.

For detailed codes, see Response codes and messages.

Get started

Use this sequence to move from provisioning to your first validated translation workflow. Keep test payloads small at first, then scale request size and concurrency after you validate authentication, endpoint routing, and response handling.

1. Create a Translator resource.
2. Get access keys and endpoint URL.
3. Run the text translation quickstart in your preferred language:
   • C#/.NET
   • Go
   • Java
   • JavaScript/Node.js
   • Python

Related content

• Translate API reference (2026-06-06)
• Migration guide (2026-06-06)
• Adaptive custom translation