Translation rates in the UK in 2026:
human, MTPE and AI pricing

Ask five translators what translation costs per word and you may get five different answers.

That does not necessarily mean somebody is overcharging. A freelance translator quoting an agency, an agency quoting its client and a specialist translator working directly for a law firm are selling different things.

And in 2026 there is another complication: “translation” might mean human translation, machine translation with post-editing, or AI-generated translation with little or no human involvement.

So a rate such as £0.10 per word tells you surprisingly little unless you also know who is charging it and what work is included.

So, what is a normal UK translation rate?

There is no official UK tariff.

Probably the best hard data we have for freelance rates comes from research carried out by the Institute of Translation and Interpreting (ITI). Its survey data was collected in 2023 and published in detail in 2024, so it should be treated as a benchmark rather than a definitive 2026 price list.

Among translators with more than ten years’ experience, the median rates reported were:

The ranges around those medians were wide. Language pair, subject matter, experience and the commercial relationship all made a difference.

Those figures are freelance translator rates, not necessarily what a translation agency charges its client.

Looking at publicly advertised UK agency prices in 2026, around £0.10–£0.18 per source word is a reasonable ballpark for straightforward professional human translation. Specialist work can cost more, but there is no sensible rule that says legal translation automatically costs X or technical translation automatically costs Y. Current published agency prices for technical and legal work commonly sit somewhere around the low-to-high teens per word rather than dramatically above them.

These are market indications, not a tariff. A quote of 8p or 20p per word can both be perfectly legitimate depending on what is being translated and what is included.

First ask: who is charging whom?

This is where a lot of online rate guides become confusing.

A freelance translator might charge an agency 8p per word. The agency might sell the same translation to its client for 14p.

The six-pence difference is not automatically profit. An agency may also be paying for project management, linguistic review, file preparation, CAT tools, QA, payment fees and revisions. It is also taking responsibility for finding a replacement if a translator disappears halfway through the project and for sorting things out if something goes wrong.

Equally, none of those things should simply be assumed to be included. Some agency workflows include independent revision by a second linguist; others do not.

If you are comparing rates, compare the service, not just the number.

Human translation is no longer the only rate you need

A translation rate card in 2026 probably needs at least three distinct services.

Human translation

A translator produces the target text and takes responsibility for the translation.

That does not mean they work without technology. Professional translators routinely use CAT tools, translation memories, terminology databases and sometimes machine-translation suggestions.

“Human translation” describes who is making the translation decisions, not whether the translator has somehow switched their computer off.

MTPE

MTPE stands for machine translation post-editing.

The source text is first translated automatically, traditionally using a neural machine-translation engine such as DeepL or Google Translate. A professional linguist then checks the output against the source and corrects it.

That can save a considerable amount of time when the machine output is good.

It can also save almost no time at all when it is bad.

This is why treating MTPE as “proofreading at half the translation rate” is a poor way to price it.

A post-editor still has to identify mistranslations, omissions, terminology errors, invented meaning, inconsistencies and sentences which are perfectly grammatical but simply say the wrong thing.

The UK’s ATC/Nimdzi pricing study found MTPE discounts ranging from effectively no discount to very large reductions depending on the amount of editing required. Recent discussions among working translators show much the same disagreement: rates around half to three-quarters of the normal translation rate are common, but they are by no means universal.

A better way to price MTPE is to ask how much faster the translator can actually work.

Suppose a translator needs to earn £45 an hour.

If they translate a particular type of text from scratch at 500 words an hour, their equivalent rate is 9p per word.

If good machine translation lets them post-edit 800 words an hour, about 5.6p per word produces the same hourly return.

But if poor MT means they manage only 400 words an hour, even 9p per word is now a pay cut.

The quality of the machine output matters more than the label attached to the job.

And what about AI translation?

“AI translation” has become an awkward term because it can mean several different things.

Modern machine-translation systems are already based on artificial intelligence, while generative AI and large language models can also be used to translate text. Some workflows combine translation memories, conventional MT, LLMs, terminology and automated QA.

From a pricing point of view, the underlying technology is less important than what happens after the machine produces the text.

There is a big difference between:

Raw AI or machine translation

The machine output is delivered without a professional linguist checking every sentence.

AI/MT translation with human post-editing

A linguist compares the machine output with the source and corrects it to an agreed standard.

Human translation

A professional translator creates and takes responsibility for the translation, whether or not software is used to assist them.

Calling all three “AI translation” makes prices impossible to compare.

Raw AI translation can be extremely cheap to produce. What costs money is the human expertise added afterwards.

For internal documents where the reader simply needs to understand the gist, raw AI may be entirely adequate. For customer-facing material, contracts, clinical information or anything where an error matters, the cost of appropriate human review quickly becomes the important part of the quote.

The sensible question for a buyer is therefore not “Do you use AI?”

It is “Who checks the final translation, and what level of quality are you committing to?”

CAT discounts are a separate thing

Machine translation and CAT-tool discounts are often lumped together, but they are not the same.

A CAT tool analyses a new project against an existing translation memory. It may find that some sentences have already been translated, some are similar to previous translations and some are completely new.

An agency can then apply a pricing grid to those categories.

For example:

There is no universal discount grid.

Nor does a 100% match necessarily mean “no work”. A translator may still need to check that an old translation makes sense in its new context, that terminology has not changed and that the segment fits grammatically with the sentences around it.

The commercial advantage of CAT analysis is not that every repeated word becomes free. It is that both sides can agree in advance how existing translation assets affect the price.

This produces a weighted word count: the commercial equivalent of the raw source word count after the agreed CAT discounts have been applied.

Per-word pricing isn’t always the right answer

Per-source-word pricing remains common for translation, particularly for European languages, because everyone knows the price before work starts.

But it is not the only sensible model.

Small certified documents are often sold for a fixed minimum fee or a price per document. Chinese and Japanese may be priced using characters rather than conventional word counts. Revision, linguistic QA and difficult MTPE are sometimes better charged by the hour.

Formatting and desktop publishing are usually separate again.

The important thing is not to force every language service into a single “price per word” field.

Other things that move the price

Two apparently similar translation jobs can still cost very different amounts.

Language pair and direction matter because translator supply differs. English into French is a different market from Finnish into English, and reversing a language pair does not necessarily produce the same rate.

Subject matter matters when genuine specialist knowledge is required. A translator who understands derivatives documentation or medical-device regulation is selling more than fluency.

The required workflow matters. Translation alone, translation plus independent revision, and MTPE are different services.

File format matters. A clean Word document is easier to process than a badly scanned PDF full of tables, stamps and handwritten annotations.

Deadline matters. Asking somebody to clear their diary for an overnight project costs more than giving them a week.

Volume can matter, although “more words” does not automatically mean they should become cheaper.

And minimum charges matter enormously on small jobs. Nobody can sustainably invoice 80 words at eight pence each once emails, file preparation, invoicing and payment administration are taken into account.

Don’t confuse translation memory with AI savings

This is becoming increasingly important for agencies.

A client might have a 50,000-word project but only 30,000 weighted words after translation-memory discounts.

That does not mean an additional MTPE discount should automatically be applied to the 30,000.

CAT leverage and MTPE describe two different forms of productivity gain. Applying both blindly can result in a price which bears little relationship to the work the translator actually has to do.

The safest approach is to define the workflow first, analyse the files and then apply the appropriate pricing model.

VAT in the UK

UK translation businesses generally quote commercial per-word rates before VAT.

If a UK supplier is VAT registered and the supply is subject to UK VAT, the standard rate is currently 20%. Cross-border work can be treated differently under the place-of-supply rules, particularly for B2B services.

So VAT should be handled separately from the linguistic rate rather than quietly folded into the price per word.

If there is any doubt about the VAT treatment of a particular client or country, check with HMRC or your accountant rather than relying on a translation rate guide.

A better way to build a translation rate card

The days when an agency could maintain one spreadsheet containing:

French: £0.12/word

are disappearing.

At a minimum, a useful rate card now needs to know:

• the source and target language;
• whether it is a client sell rate or supplier buy rate;
• the service: human translation, revision, MTPE or another workflow;
• the specialism, where relevant;
• any CAT-tool discount grid;
• minimum charges;
• urgency rules; and
• additional services such as certification or DTP.

For MTPE in particular, it is worth measuring actual productivity rather than deciding that every machine-translated project receives the same percentage discount.

That also makes margin much easier to understand. If your client rate is 14p and your translator rate is 8p, that is useful information. If half the project consists of discounted matches, the translator is working at an MTPE rate and somebody has manually copied the analysis figures into three different systems, the headline rates tell you very little about whether the project actually made money.

This is one reason we built rate cards and CAT analysis into VectorLingo: the client rate, supplier cost, weighted word count, project and eventual invoice should be working from the same numbers rather than being re-keyed along the way.

The short version

If you simply need a starting point, around £0.10–£0.18 per source word is a reasonable published UK client-price range for straightforward professional human translation in 2026, with plenty of legitimate work falling outside it. Freelancer rates paid by agencies are generally lower.

But the more useful lesson is that there is no longer one thing called a “translation rate”.

Human translation, MTPE and raw AI translation make very different promises to the buyer and involve very different amounts of professional work.

Price those promises separately.

Sources

• Institute of Translation and Interpreting, Freelance Translator Rates: The Value of Experience (published 2024; survey data collected in 2023).
• Association of Translation Companies / Nimdzi Insights, UK Language Services Pricing Study (2020), including MTPE pricing.
• Current publicly advertised UK agency prices were also reviewed to sense-check client-facing 2026 ranges. These are advertised prices rather than independent market statistics.