Contents6
In short
Human translation is paid per page or per word; the more publishable the text must be, the closer the price gets to a full translation, even when starting from machine output.
Machine translation is paid per character or per token, and its bill stays small at the scale of a website, whatever the tool.
The cost that decides is the market’s: translation, review, research in the language, customer service and updates, over a year.
“How much will it cost to translate our website?” The question calls for a number, and every provider gives one, counted in its own way: per word, per page, per character, as a flat fee. Quotes become impossible to compare and, above all, none of them says what the language will cost once it is open, when it has to be kept up to date and the customers who speak it have to be answered.
This article puts the three ways of counting side by side, with public prices, then adds what quotes leave out. It extends our articles on website translation and opening new markets.
Per page or per word: what human translation pays for
A translation done or reviewed by a person is billed on the volume of source text: per word, most often, or per page. Market rates vary widely by language, field and turnaround, and few price lists are published. One has the advantage of being official and detailed: that of the Translation Centre for the Bodies of the European Union, which translates for EU agencies and publishes its prices every year in its budget decision for 2026.
Price per page at the EU Translation Centre in 2026, standard turnaround
This is not a market price: it is the price list of a public body billing its institutional clients. But its structure says what matters.
The more publishable the text must be, the closer the price gets to a translation. Full post-editing, which reworks machine output until it is comparable to a human translation, costs almost as much as a translation there. Light post-editing costs less, but the Centre reserves it for texts that are not intended for publication. A website is, by definition, published.
The machine alone costs almost nothing. Thirty cents a page, about three hundred times less than a translation. The gap does not read as “the machine replaces the translator”: it measures the work a person still does so that a text is accurate, in the right tone, with the right terms.
Turnaround and format are paid for. In the same price list, urgent translation costs 113.50 € a page, a quarter more than the standard turnaround, and a complex format adds 15 € per page and per language.
Per character: what machine translation costs
Machine translation services and general-purpose AI models are billed on the volume processed, but each counts in its own way. Four public price lists, checked on 19 October 2026:
per million characters with the translation model of Google Cloud Translation, beyond 500,000 characters included each month
a month at DeepL, billed annually, for 300,000 characters a month and one glossary
per million input and output tokens for Gemini 3.1 Flash-Lite, a model Google describes as optimised, among other things, for translation
per million input and output tokens for GPT-5 mini, at OpenAI’s standard rate
The units do not compare directly. A token is about four characters for Gemini models, according to Google’s documentation on tokens, each provider splits text in its own way, and an AI model also bills the text it produces, as well as the instructions and glossary sent with each batch. The DeepL plan quoted is the app’s, per user; API access has its own price list. And Google also offers, within Cloud Translation, an LLM-type translation model billed at $10 per million input characters and $10 per million output characters.
What these price lists say together: at the scale of a website, the machine’s bill stays low, whatever the tool. They say nothing about quality on your texts, which is measured separately. We compare the tools, with no favourite, in DeepL, Google or an LLM: which translates your trade best.
Per market: the cost that decides
The price of a translation is only the first line. An open language is a market to serve, and it is the sum of everything it requires that decides whether it pays off.
- The initial translation.
The pages, but also menus, buttons, forms, automatic emails and title tags.
- The review.
In full for the pages that commit the company (offers, prices, terms), on template and sample for repetitive pages: we detail that split in translating your website with AI, automatic or reviewed.
- Research in the language.
Keywords are not translated, they are researched locally, and URLs and hreflang tags are set once and for all: see the hreflang mistakes that cost the most.
- Adapting to the country.
Prices, formats, payment methods, local proof: that is the subject of localize, don’t just translate.
- Service in the language.
Someone answers the requests, quotes and complaints that arrive in that language.
- Updates.
Every modified page goes back to translation and review, for the whole life of the site.
The last line weighs most over time, and it is the only one that can be reduced structurally: a translation memory only sends back what has changed, and a brand glossary avoids paying twice to correct the same term.
From the field
On Palmora Property, a real-estate listings site in six languages that we operate, changing translation provider cut the cost of the pipeline by a factor of 60, and 95% of segments come out of the cache after the first pass, without calling the AI again. German was opened through the blog alone first: 1,092 content keys in parity with the other languages, 24 articles, 2,009 URLs, a test of demand before translating the rest. These figures hold for that site: the share reused from the cache depends on what changes from one update to the next.
What makes a translation quote vary
For the same volume, two quotes can differ widely. The factors that explain the gap:
- The language. A rare language, or one outside the EU, costs more: the Translation Centre adds 10 € per page for a non-EU language.
- The field. A legal, medical or technical text needs a translator from that field, and their responsibility is paid for.
- Repetitions. A catalogue where the same sentences come back costs less with a memory: ask how the quote treats segments already translated. For a catalogue, see translating product pages for export.
- The format. A text delivered in a spreadsheet or an export file is translated faster than one that has to be extracted from pages or PDFs.
- The turnaround. Urgency is paid for, everywhere.
- The level of review. A full review, a sample review or no review give neither the same price nor the same risk.
Key takeaway
The right question is not “how much does translation cost”
It is “how much does an open language cost over a year”, translation, review, research, service and updates included. A market that does not cover that total does not become profitable with a cheaper translation.
Where to start
Do the sums for a single language, that of your first market, and check first that there is demand: our method for choosing markets before translating starts from the searches already visible in Search Console. For an online store, the order of the work is set out in multilingual and multi-currency store: where to start.
To price a full language, translation, review and maintenance included, describe your site and the target market: we start from the real volume of your pages and what changes each month.
Frequently asked questions
How much does it cost to translate a web page?
There is no single price: it depends on the length of the page, the language, the field and the level of review. As a public benchmark, the EU Translation Centre charges 91.50 € per translated page at standard turnaround in 2026. Market providers set their own rates, most often per word: compare them at equal volume and equal review.
Is machine translation enough for a website?
To understand a text, often; to publish it, rarely without review. The machine costs little, but an error on a page that commits the company costs more than its review. The right split is made page by page: full review of what commits you, sample checks on what repeats.
Why does an update cost less than a first translation?
Because with a translation memory, only new or modified sentences go back to translation. Without a memory, each update may be retranslated in full, or require someone to spot by hand what has changed. Ask the provider how it bills sentences already translated.
Do you still need a translator if you use AI?
Yes. AI produces a fast, inexpensive first draft; a translator or reviewer who masters the language remains necessary for the pages that sell or commit the company, to fix the trade’s terms and to judge the tone. Their work shifts towards review, terminology and quality control; it does not disappear.
Sources
- Translation Centre for the Bodies of the EU, 2026 budget (applicable prices) (new tab)cdt.europa.eu
- Google Cloud, Cloud Translation pricing (new tab)cloud.google.com
- DeepL, Translator plans (new tab)deepl.com
- Google, Gemini API pricing (new tab)ai.google.dev
- OpenAI, API pricing (new tab)developers.openai.com
- Google, understand and count tokens (new tab)ai.google.dev
Translation Centre for the Bodies of the EU, 2026 budget (applicable prices)
Read on cdt.europa.eu(new tab)Jim · founder of Brain Plus AI
Also in translation and new markets
Choose and price7 min
Translating your website with AI: automatic or reviewed?
Having everything reviewed is expensive, and reviewing nothing lets errors through in languages that nobody at your company reads. The split is decided page by page, depending on what each page commits you to.
Understand7 min
Translation memory: what a company gains from it
A sentence translated once should not be paid for twice. The translation memory makes sure of that at every site update, provided it belongs to you.
Choose and price7 min
DeepL, Google or an LLM: which translates your trade best
Every provider publishes a test it wins, and independent evaluations crown a different winner from one year to the next. For a specialised text, the only reliable answer comes from a trial on your own texts, language by language, judged by someone who reads the language.