MTPE Rates per Word: How Translators and Agencies Price Post-Editing

Published MTPE rates per word disagree, but a workable start is light post-editing at about 40% to 65% of your human per-word rate and full post-editing at 50% to 70%, adjusted after a sample of the machine output. Quote per source word, name the quality tier, set a minimum fee and say what happens if the machine output turns out poor.

Percentages are a shortcut for effort. A skilled editor working on poor machine output can take as long as translating from scratch, so a flat 50% can quietly pay less than the work costs. Below are the published bands with their sources, then an effort-based sum that tests any percentage before it goes into a quote.

Follow me on Instagram@sagnikteaches

A word on the evidence. Every figure here comes from an agency's own pages, an analyst article or a survey run by an agency. None is an independent market survey, and rates move with language pair, subject and client. Each page was rechecked on 4 November 2026; the analyst article itself dates from 2020. Use the bands as ranges to test, not rates to copy.

Connect on LinkedInSagnik Bhattacharya

Dollar figures are USD per source word, as the vendors published them. Percentages carry across currencies, which is one reason to quote them first.

Subscribe on YouTube@sagnikteaches

Published MTPE bands, with who said what

Four sources give usable numbers. How they differ matters as much as what they say.

Turn the dollar bands into shares and the disagreement shows. Taking midpoints, Tomedes's light band ($0.055) is about 24% of the midpoint of its human band ($0.225), and its full band ($0.10) is about 44%; that is my arithmetic, not the page's. ICS Translate puts light at 40-50% and full at 50-70%. Nimdzi's "up to 35-40% off" for light work implies paying 60-65% of the regular rate, and the survey suggests many freelancers discount little or not at all.

That is a spread from about a quarter of the human rate to nearly all of it. It looks like a contradiction but mostly reflects who is speaking: an agency pricing a client's project, an analyst thinking about fair effort, and freelancers deciding what they can accept.

Why the numbers differ, and which one is yours

Six things move an MTPE rate, and the ICS Translate page names most of them: source text quality, machine output quality, language pair, sector, volume and the number of editing passes. Add one more that the published bands ignore, which is who sits on each side of the price.

• Client price versus editor pay. An agency charging a client $0.10 a word for post-editing is not paying its editor $0.10. Compare like with like before you use a band.
• What "light" means. ICS Translate describes light output as understandable, accurate enough and free of major errors, suited to internal and lower-risk content where speed matters more than polish. A client who expects polish at the light price is buying full post-editing at a discount.
• Machine output quality for your pair and subject. This is the largest swing factor, and the only one you can measure in ten minutes.
• Risk of the content. Marketing copy, legal text and safety instructions justify the full tier or more, whatever the machine produced.
• Who supplies the machine output and tools. If you run the engine and carry the subscription, that cost belongs in your rate.
• Repetition. Text that matches earlier translations is usually priced by a weighted grid; see translation memory and AI in a small agency for how the two discounts combine without double-counting.

The bands in dollars at three human rates

Percentages need a base. This table applies the light band (40-65%) and the full band (50-70%) to three human rates. It is arithmetic on the published percentages, so it inherits all their limits.

Where you sit inside each range is the sample test's job, covered below. Clean output and a tolerant client point to the bottom; poor output or risky content point to the top. If a sample pushes you above the range, treat that as information too, because the file may be better translated from scratch.

For comparison, Tomedes's $0.03 to $0.08 for light work reaches below the $0.16 row, which suggests its light price assumes cleaner output or lower human rates, while its $0.05 to $0.15 for full post-editing brackets that row. Those are one agency's price points across many pairs. A specialist in a sector with scarce linguists may reasonably sit higher.

Pricing by effort: a $320 day and four scenarios

Start with what you need to earn in a working day; take $320 as the figure. Divide it by the words you can handle in that day and you have a per-word rate that already contains your effort. Tomedes's FAQ puts a single human translator at 2,000 to 3,000 words a day and says an MTPE workflow can bring a 20,000-word project down to 2 to 4 days; swap in your own pace.

The first three rows land inside the published bands. The fourth shows why a flat percentage is dangerous: on poor output the same editor needs more than the human rate to earn the same day. Nimdzi makes a related point about hourly quotes, warning against dividing a project by an assumed productivity gain. The remedy is to measure your own pace, not to assume one, and to write a quality condition into the quote.

A 500-word sample test before you quote

Ten minutes on the real file beats any published band. This routine turns "it depends" into a number you can defend.

1. Pick 500 words that represent the whole file, including the hard paragraphs, not only the easy opening.
2. Run them through the machine engine the project will use.
3. Post-edit at the tier you intend to quote, and time yourself.
4. Convert the time to words an hour, then to words a day, counting only the productive hours.
5. Divide your day rate by words a day, and round the result up.
6. Note the share of segments you had to rewrite, since that figure sets the threshold in the quote.

A small manufacturer sends 40,000 words of product manuals. The translator's sample of 500 words takes 40 minutes at full post-editing, which is 750 words an hour. With five productive hours a day, the rest going on queries, terminology and admin, that is 3,750 words a day, so $320 divided by 3,750 gives $0.085 a word, about 53% of a $0.16 human rate. The job quotes at $3,400 against $6,400 for human translation.

A second client sends a file in a different pair and the same test takes 75 minutes: 400 words an hour, 2,000 a day, and $0.16 a word, the full human rate. Same translator, same day rate, nearly double the per-word price. A single "MTPE rate" on the website would have underpaid the second job or overcharged the first.

Three quality tiers to put in the quote

Naming the tier stops "post-editing" meaning whatever the client hopes it means. The price positions below are ranges from the sources above, not rules.

ISO 18587:2017 sets requirements for post-editing machine translation output, and vendor pages describe full post-editing as the level it defines. A client who asks for "ISO-compliant post-editing" usually means the full tier, so confirm which they expect; ISO 18587 certification for a small translation agency covers whether certification is worth having. To help a client choose between light and full, is machine translation good enough for client-facing documents gives a plain test.

Quote wording you can adapt

A good quote answers the client's questions before they ask them. This template fits a freelancer and, with the names changed, an agency.

Quote: post-editing of machine translation, [source language] into [target language]
Scope: [word count] source words from [file name]. Machine output supplied by [client / us] using [engine].
Tier: Full post-editing. Meaning, grammar, terminology (client glossary applied), tone and consistency, to publishable quality.
Rate: $[x] per source word. Minimum fee $[y].
Sample: 500 words post-edited free before this rate is confirmed.
Quality condition: this rate assumes no more than [z]% of segments need a full rewrite. Beyond that, the rate moves to $[higher] or the job reverts to human translation at $[human rate].
Excluded: terminology research beyond the glossary, desktop publishing, back-translation, second-linguist review.
Revisions: one round of client feedback within 14 days of delivery.

Three clauses do most of the protecting. The quality condition makes the rate fair when the sample was kinder than the real file. The exclusions stop "post-editing" absorbing review and layout work. The minimum fee covers the fixed cost of opening a file, which a 300-word job otherwise can't: at $0.085 a word, a $40 minimum covers 470 words, so anything shorter bills at the minimum. Before you send a quote, check how client text reaches the engine: putting client documents into AI translation tools lists the confidentiality questions to ask first.

A mixed project priced as three lines

Few files are one content type. Take the manufacturer's 40,000 words again, now split three ways: 25,000 words of product manuals, 10,000 of support articles and 5,000 of marketing copy. A translator with a $0.16 human rate quotes the manuals at full post-editing, $0.09 a word, which is $2,250. The support articles go at light post-editing, $0.064 a word, which is $640. The marketing copy, where persuasion matters more than speed, stays human translation at $0.16, which is $800.

The total is $3,690, a blended $0.092 a word and about 58% of the human rate, against $6,400 if everything were translated from scratch. The client sees three lines with three reasons instead of one discount, which makes it harder to negotiate down the line that carries the most risk. If the manuals sample comes back poor, only that line moves.

When a client says it is only editing, so half price

The expectation is common. In the agency-run survey above, about half of respondents said AI and MTPE had significantly shaped what clients expect to pay. Answer it with options rather than a refusal.

"Hi [first name], thanks for the file. The machine output is a useful starting point and I've priced to reflect it: light post-editing at $[x] a word if you need to understand the content, full post-editing at $[y] if it will be published. The rate depends on how much of the draft I can keep, so I'd like to post-edit 500 words of the real file first, free of charge, and confirm the rate on the result. If the rest of the file is poorer than the sample, the quoted rate holds up to [z]% rewritten segments; beyond that we'd talk before I carry on."

That reply does three things. It offers two tiers, so the client chooses the quality it needs and the price follows. It turns the argument into evidence, using the sample. And it caps your exposure without sounding defensive. If the client still insists on a flat discount, offer hourly work with a cap instead, but only if you can log your time honestly and the client accepts the log.

Agencies: building the editor's rate and the client price

An agency has two numbers to set: what it pays the editor and what it charges the client. Use the sample test for both, then build the client price from cost. Consider an agency pricing full post-editing for a 10,000-word manual. The editor is paid $0.08 a word ($800). A second linguist spot-checks 20% of the words at $0.03 per checked word (2,000 x $0.03 = $60). Project management and tools add $90. Total cost is $950, or $0.095 a word.

For a 30% gross margin on price, divide the cost by 0.7, not multiply by 1.3. That gives $1,357 for the job, or about $0.136 a word. Multiplying by 1.3 gives $1,235, which is only a 23% margin. The gap is $122 on one small project, and it grows with every job.

Paying editors a flat share of their normal rate is simple but ignores output quality. A fairer arrangement pays the tier rate on clean output and a pre-agreed higher rate when the sample or the file shows a heavy rewrite. The translator's post-editing workflow shows where in the job to take that measurement, and what AI translation versus human translators really cost gives the client-side view you will be negotiating against.

Publishing rates on your website without anchoring yourself

A rate on a website becomes the starting point of every negotiation, so publish it with care. Show a "from" figure for each tier, state the condition, and point to the sample: "Light post-editing from $0.06 a word. Full post-editing from $0.09. Final rate confirmed after a free 500-word sample." That wording invites the sample rather than the argument.

• Don't publish a single MTPE number: one figure can't be right for both clean and poor output.
• Don't publish a rate without the tier definition beside it, or "light" will be read as "full, but cheaper".
• Do state the minimum fee, so small files don't arrive as a surprise.

Checking the rate after three jobs

A rate card is a hypothesis until it has met real files. After three jobs, compare quote with reality.

• Effective hourly rate: fee divided by the hours you actually spent, against your target.
• Hours quoted by your sample versus hours used, per job.
• Share of segments rewritten, against the threshold in the quote.
• Client revision requests, and whether they point to a tier mismatch.

If your effective rate falls short of target by more than about 15% on two of the three jobs, raise the band for that pair and engine, or tighten the quality condition. If the effective rate beats the target comfortably and clients accept without comment, resist the urge to cut: put the extra margin towards a second review for the high-risk text.

MTPE pricing questions translators and agencies ask

Should MTPE be priced per word or per hour?

Per source word is the most common basis, according to a TAUS survey cited by Nimdzi, and it works well when the quote carries a quality condition. Use an hourly rate with a cap when content or machine output is unpredictable and you can track your time. Avoid hourly rates built by dividing the project by a guessed productivity gain.

Is a 50% discount for post-editing normal?

Sources vary. One agency-run survey found about half of freelance respondents give no MTPE discount, and among those who do, 10-30% was the most common range. One agency page quotes light work at 40-50% of the human rate. A flat half price for full post-editing can fall below the cost of the work when the machine output is poor.

What if the machine output is too poor to edit?

Put a threshold in the quote, based on your sample, such as the share of segments needing a full rewrite, and say what happens beyond it: reprice at the next tier or revert to human translation at your usual rate. Raise it before you start, with examples from the sample, not after the hours are spent.

Do I charge the same MTPE rate for every language pair?

No. Machine output quality differs by pair, subject and engine, so test each combination with a sample. A rate that works for a widely supported pair can undercut you on a rarer one, where editing may take nearly as long as translating from scratch.

Should I tell clients when machine translation is used?

Yes, in the quote and the agreement. Say which tier you are quoting, who supplies the machine output and which engine or tool processes the text. Check the client's confidentiality terms before any of their text goes into an online engine.

Further reads

• Subtitling Rates per Minute: What to Charge When AI Drafts First — The same repricing question for subtitle work
• DeepL vs ChatGPT for Translation: Which to Use for Each Client Document Type — Test which engine gives you the cleanest draft to edit
• How Translators Use AI to Keep Terminology Consistent — Keep terminology steady across a post-editing job
• How Small Translation Agencies Build an AI-Assisted Workflow — Build the workflow the rate card sits on
• Does Your Translation Business Need an AI Usage Policy? — Write down when and how your business uses machine output
• memoQ vs Trados for Small Translation Agencies: AI Features and Cost — Compare tools that track post-editing effort
• AI Subtitle Translation for Agencies: Quality Checks Before Delivery — A cue-by-cue workflow for AI subtitle translation, with timing, line-length and meaning checks, a QC sample and a brief for the native reviewer.
• AI Legal Translation: What a Certified Translator Can Use and Disclose — Certification rules differ by country; this explains AI as a draft only, the translator's declaration and what to tell a client.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: ICS Translate (August 2026), Tomedes cost page (updated September 2026), Nimdzi (February 2020) and GTS Translation's survey on its own website (April 2025), rechecked on 4 November 2026. Day-rate and agency figures are illustrative.