Translating with artificial intelligence seems almost free, but how much does a publishable-quality AI translation really cost? The price of machine translation is only part of it: the real cost depends mostly on post-editing, the human review needed to make the text correct and fluent. This AI translation cost calculator combines machine translation and review and tells you the figure that matters: the real cost per word, the total monthly cost and the saving versus a translator or an agency.

What the real cost of AI translation is

When people talk about "free AI translation" they mean only the machine translation (MT), which today costs very little. But raw machine translation is rarely publication-ready: it contains inaccuracies, terminology errors and unnatural passages. To use it in a professional context you need post-editing, a human review that has a real cost, usually far higher than the machine translation itself. The true cost of AI translation is therefore the sum of the two: machine plus review.

How the calculator works

Enter five values: the words to translate per month, the machine translation cost per 1000 words, the post-editing minutes per 1000 words, the reviewer's hourly cost and the price of a human translation per word. The calculator shows the machine translation cost, the review cost, the total AI cost, the cost per word and the comparison with human translation, with an instant verdict. All in real time: change a value and see the effect at once.

The five values you need

The words per month define the volume; the machine translation cost is the spend for the machine (low today); the post-editing minutes and the reviewer's hourly cost determine the review cost, by far the heaviest item; the human price per word drives the comparison. Change the review minutes and you instantly see how much they weigh: it is the key variable of the AI translation cost.

Machine translation plus post-editing: the real cost

The surprise, for those who miscalculate, is the cost breakdown: the machine costs very little, the human review almost everything.

That is why the real saving of AI translation does not come from zeroing the machine cost, but from reducing the review time: a good machine translation needs less post-editing and therefore costs less.

What post-editing (MTPE) is and why it costs

Machine Translation Post-Editing (MTPE) is the work of a human translator who reviews the machine's output: fixing errors, sorting out terminology, improving style and ensuring the meaning is faithful. It is skilled work, paid by time or by word, and it is why AI translation is not free. How much post-editing is needed depends on the machine's quality, the language pair and the required quality level.

Post-editing levels: light and full

Not all review is the same. There are two main levels, with different times and costs:

Light post-editing is enough for internal or informational content; full post-editing is needed for public, polished text. Choose the minutes in the calculator based on the level you need.

How to estimate the review minutes

It is the trickiest input. An expert reviewer can post-edit from 1,000 to 4,000 words an hour depending on the machine quality and the required level: that means about 15-60 minutes per 1000 words. If you have no data of your own, start from a cautious estimate (30-40 minutes for full post-editing) and refine it by measuring the real time on a sample. It is the variable to be realistic about, because it drives most of the cost.

The cost per word: the metric that counts

The cost per word is the figure that makes AI and human translation comparable. Raw machine translation can cost fractions of a cent per word, but with post-editing the real cost rises to a few cents; a professional human translation typically costs from €0.08 to €0.20 a word or more, depending on the language pair and sector. The calculator highlights exactly this: how much a word translated with AI really costs you, post-editing included.

AI or human translator: when which is best

The choice depends on the type of content and what is at stake:

For large, repetitive volumes AI with post-editing is almost always cheaper; for high-risk or creative texts, the human remains irreplaceable.

The quality of machine translation today

Modern neural machine translation systems and large language models have raised quality a lot: for many common language pairs the output is often good and needs little post-editing. However, problems remain on specialist terminology, ambiguity, cultural references and consistency over long texts. The better the machine on your language pair, the less post-editing you need: enter review minutes consistent with the real quality you observe.

Language pairs: not all perform the same

Machine translation quality varies a lot from one language pair to another. Very common combinations (for example English into the major European languages) have better output and need less post-editing; less-represented or structurally very different languages need more review and therefore cost more. Account for this by adjusting the post-editing minutes: the same AI can be very cheap on one pair and costly on another.

Sectors where AI is not enough

In some fields a translation error has serious consequences: legal texts (contracts, patents), medical (leaflets, reports), financial and safety. Here machine translation can be a starting point, but expert review or human translation is always needed, with professional liability. Creative marketing, where tone and wordplay matter, also rarely settles for AI. For these cases, raise the review level or choose the human.

CAT tools, glossaries and translation memories

Professional translation uses CAT (Computer-Assisted Translation) tools with translation memories and glossaries: they reuse already-translated sentences and enforce the correct terminology. Integrated with AI, they greatly reduce post-editing because the machine starts from content consistent with your past work. Investing in glossaries and memories lowers the review minutes over time, and thus the cost per word: a factor the calculator reflects if you reduce the post-editing minutes.

When AI is NOT worth it

AI translation loses its edge when the post-editing needed is very high and the volume is low: few texts, a difficult language pair and a full review can make AI cost as much as or more than a human translation, as the calculator flags with a zero or negative saving. Even for short but critical texts, the review time to ensure quality can wipe out the advantage. In these cases, going straight to the human is simpler and safer.

The hidden costs of AI translation

• Project management: preparing files, setting up flows and checking deliveries has a cost.
• Final quality control: beyond post-editing, a final check (QA) is often needed.
• Layout and formats: adapting the translation to layouts, subtitles or software adds work.
• Data confidentiality: for sensitive texts you may need a secure or on-premise tool, which costs more.

You can account for these by raising the review minutes or the hourly cost in the calculator.

How to reduce the translation cost

• Use good-quality machine translation on your language pair: less post-editing, less cost.
• Build glossaries and translation memories to reduce repeated corrections.
• Choose the right post-editing level for each content, without overdoing it where it is not needed.
• Write the source texts clearly and in standard language: the machine translates them better.

Common mistakes in calculating the cost

• Considering only the machine translation cost and ignoring post-editing.
• Underestimating the review minutes needed for publishable quality.
• Using the same review level for all content, even critical ones.
• Comparing AI with human without considering the difference in quality and liability.
• Forgetting management, QA and layout costs.

How to start with a pilot

The most reliable way to fill in the calculator is a small pilot: take a representative sample of your texts, translate it with AI and time the real post-editing needed to bring it to the quality you require. You will get fact-based review minutes to enter here for a credible estimate, before setting the flow on all your volumes.

Practical example

A company has to translate 50,000 words a month. Machine translation costs €0.20 per 1000 words, full post-editing takes 30 minutes per 1000 words and the reviewer costs €35 an hour. The result: €10 of machine translation, €875 of review, for a total of €885, that is about €0.018 a word. A human translation at €0.10 a word would cost €5,000: AI with post-editing saves over €4,000 a month and is "very worthwhile". Enter your numbers for the exact estimate.

Updated July 2026