The billing policy for Azure AI Translator custom training.

Hello Polonez Caro,

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that you would like to know details about the billing policy for Azure AI Translator custom training.

Regarding your questions, let's break it down:

Can the $200 Azure Free Account credit be used to pay for Custom Translator model training?

Yes. The $200 Azure Free Account credit (available for the first 30 days after signup) can be applied toward Custom Translator model training costs.

Custom training is considered a standard billable Azure resource. Since the free credit applies to most billable Azure services unless explicitly excluded, Custom Translator training charges are eligible.

However:

• The credit expires after 30 days.
• Once the $200 is consumed, services stop unless you upgrade to Pay-As-You-Go

Azure Translator Pricing:
https://azure.microsoft.com/pricing/details/cognitive-services/translator

Can the Translator F0 (Free tier) allowance be used for custom model training?

No. The Translator F0 free tier does not cover custom model training.

The F0 tier only provides a limited number of free characters per month for standard text translation API usage. It does not include:

Custom model training

Custom model hosting

Custom category inference

These are billed separately. Azure AI Translator Pricing (Custom Model Section):
https://azure.microsoft.com/pricing/details/cognitive-services/translator

If I train a custom model but do not deploy it, can I leave it undeployed for several months without being charged?

Yes. Once training completes, you are charged only for the training process itself (based on compute hours used during training).

If the model is not deployed:

There is no hosting charge.

There is no hourly fee.

The model remains stored in your workspace.

You can leave it undeployed for months without incurring hosting costs. For clarification, charges apply only when:

• The model is actively deployed.
• It is processing translation requests.

Custom Translator Pricing – Training vs Hosting:
https://learn.microsoft.com/azure/ai-services/translator/custom-translator/overview

If I upgrade later from Free to Pay-As-You-Go, will my trained models still exist?

Yes, your trained models will remain available provided:

The Azure subscription is still active.

The resource has not been deleted.

You upgrade before the subscription is disabled.

Upgrading from a free account to Pay-As-You-Go does not delete:

Workspaces

Uploaded datasets

Trained models

The risk scenario: If the subscription expires and is not upgraded in time, Azure may suspend or delete resources according to subscription lifecycle policies.

Azure Subscription Lifecycle:
https://learn.microsoft.com/azure/cost-management-billing/manage/subscription-lifecycle

Is custom model training billed based on the number of source and target characters?

No. Custom model training is not billed per character.

Training is billed based on compute time (measured in hours) used during the model training process.

However:

• Translation usage after deployment is billed per character translated.
• Hosting is billed per hour while the model is deployed.

Azure Translator Pricing Page: https://azure.microsoft.com/pricing/details/cognitive-services/translator

For my use case (Japanese → English → Chinese), how will billing apply?

Your workflow:

• Japanese > English
• English > Chinese
• Used later in Japanese > Chinese workflow

The implication:

1. If you translate in two steps (JA > EN > ZH), you are billed for characters twice.
2. Each language pair requires its own custom model training (if using custom models).
3. Each deployed model incurs separate hosting charges.
4. Translation usage is charged per character processed per step.

If you translate 1 million Japanese characters:

• JA > EN = 1 million characters billed
• EN > ZH = 1 million characters billed again
• Total billed = 2 million characters

If possible, training a direct JA > ZH custom model may:

• Reduce translation cost (single processing step)
• Reduce quality loss from multi-hop translation
• Simplify deployment

I hope this is helpful! Do not hesitate to let me know if you have any other questions or clarifications.

Please don't forget to close up the thread here by upvoting and accept it as an answer if it is helpful.