Multilingual social software: Amazon Translate and Google AutoML text costs

Match the price to the translation operation

For multilingual social copy, price the operation rather than the provider name. Amazon Translate Standard Text Translation, Google AutoML Translation text and Azure custom translation have different published rates. Do not treat a standard text charge, a custom translation charge and a document charge as interchangeable.

Amazon Translate's Standard Text Translation costs $15.00 per million characters, with a free tier of 2 million characters per month for 12 months. The Free Tier applies to some translation services and starts from the date of the first translation request.

Published text and document charges

Keep text and document charges separate. Azure prices are calculated based on US dollars. For Google, prices in currencies other than USD are listed on Cloud Platform SKUs. Retain the published dollar amounts when budgeting, rather than converting them into pounds without a stated exchange rate.

Google document pages are counted by page breaks or slides. Azure prices are estimates, not actual price quotes, and can vary with the agreement, purchase date and currency exchange rate. Keep human editing separate when comparing translation quotes for a social calendar.

Free allowances and character counting

Azure F0 - Free includes 2 million characters of any combination of standard translation and custom translation training free per month. If you intend to share this allowance between tasks, calculate the remaining balance before sending the month's social calendar for translation. Keep training usage visible in that calculation.

The Microsoft Translator free subscription stops if it reaches 2 million characters during a subscription month. Service resumes at the beginning of the next subscription month or after changing to a paid plan. Switching to a paid subscription starts billing with the first character after the switch.

Amazon Translate includes white space characters. Google counts characters sent for processing, including whitespace and HTML tags. Microsoft Translator bills text translation by input characters; each target language and repeated translation counts separately. For an estimate, multiply the source character count by the number of target languages.

Illustrative monthly social-calendar costs

Illustrative example: assume a monthly social calendar containing 500,000 source characters for translation into two target languages. The calculation is 500,000 x 2 = 1,000,000 translated characters. Assume no other account usage and that the relevant free allowances apply. These inputs describe an example, not a forecast.

These calculations cover the named text operations, not equivalent service configurations or a quality ranking. They exclude human editing, document work, training and hosting. Use the relevant operation's charge within social project costs, but keep the translation subtotal distinct from the budget needed to prepare and publish approved copy.

Customisation and data handling

Amazon Translate can customise batch output at runtime using parallel-data translation examples included with a job request. Parallel-data storage includes 200 GB free per account; excess storage costs $0.023 per GB per month. Keep this storage charge separate from translation volume when costing customised batch work.

Microsoft Custom Translator can build a custom language model from pre-existing translated text to reflect a domain, terminology or style. Azure Custom Translation model hosting carries a flat fee for every model hosted during a billing period. That fee is not pro-rated for hosting lasting less than the full month.

Google AutoML Training Operations cost $45 per hour, with a maximum of $300 per training. Keep training totals separate from text charges in translation workflows. A glossary for human reviewers should also remain distinct from a trained model, even where both are intended to support consistent terminology.

Customer data submitted to Microsoft Translator for translation is not written to persistent storage, and there will be no record of the submitted text in a Microsoft data centre. Treat that statement as specific to submitted translation text; check uploaded training examples and stored resources separately.

Selection checks for social copy

Choose the operation that matches the content, then separate language judgement from its price. Treat AI applications as bounded tasks with a human decision before publication. Ask whether terminology, document formatting or data handling changes the operation you need before using a low translation subtotal to justify selection.