MTPE Pricing Models: Per-Word vs Hourly vs Effort-Based Compared

You send a post-editor a 30,000-word project at $0.06/word because the client said “the MT is pretty good, just needs a light pass.” Three days in, the post-editor emails back: “This output is rough. I’ve been retranslating segments for two days. I can’t finish this at LPE rates.” Meanwhile the client already expects the LPE price you quoted.

That’s not a rare scenario - it’s the most common breakdown point in MTPE projects. And almost always, it traces back to a pricing model that wasn’t right for the situation.

Here’s a head-to-head breakdown of the three models, what each one actually costs, and how to pick the right one before you’re stuck in the middle.

The three models - a quick orientation¶

Every MTPE project gets priced one of three ways:

Per-source-word - you charge a flat rate per word, applying a discount to your standard human translation rate. $0.20/word human rate becomes $0.07/word for LPE, $0.12/word for FPE. Simple, scalable, comparable.

Hourly - you charge by time spent. Client-facing rates typically run $35-$85/hour. Translator-side: $20-$50/hour depending on content type. Protects against MT quality surprises. Clients hate the open-endedness.

Effort-based (TER/editing distance) - payment is linked to how much the post-editor actually changes the MT output, measured by Translation Edit Rate. Segments touched minimally get reduced rates; heavily rewritten segments get full rate. Theoretically the most accurate reflection of actual work. Practically the hardest to implement.

Most agencies live in model one and call model two a backup. Model three is used mainly by large LSPs with dedicated tooling and established translator relationships.

Per-word: the industry default¶

Per-source-word pricing is what clients expect to see, what’s easiest to compare across agencies, and what scales cleanly with volume. Nimdzi’s survey found 69.6% of translators prefer to be paid on a per-source-word basis. It’s the standard for good reasons.

The current market rates (2026):

Source: Artlangs MTPE Pricing Guide 2026, Weglot analysis.

One thing that jumps out from these numbers: daily earnings for LPE and FPE editors often converge. An LPE editor doing 5,000 words at $0.04/word earns $200/day. An FPE editor doing 2,000 words at $0.09/word earns $180/day. The rate-per-word difference compensates for the throughput difference. If you’re paying significantly less per word for FPE than LPE, your model has a problem.

When per-word works well: - MT quality is known from prior projects or a pilot - Content type and domain are consistent (same language pair, same subject matter) - Volume is high enough that per-word scales better than hourly (typically 10,000+ words) - The client wants a fixed project cost before work starts

When per-word breaks down: - First project with a new MT engine or language pair (MT quality is unknown) - Heterogeneous content where segments vary wildly in post-editing effort - Poor MT output that makes the discounted rate economically irrational

The core risk: a source word is a source word whether the MT gave you a clean output or garbage. If MT quality varies 40% across a project, your margin on per-word pricing swings by the same amount. Per-word makes sense only when you’ve validated MT quality upfront.

As a Weglot analysis puts it: poor machine translation output can push editorial time to 70-100% of standard human translation effort - at which point the per-word rate no longer reflects the actual work.

Hourly: the escape valve¶

Hourly billing is what you reach for when per-word is too risky. It makes MT quality variance someone else’s problem - because you’re billing time, not output.

Current rates:

At LPE throughput of 5,000 words/day (~625 words/hour), a translator billing at $25/hour earns the same as billing $0.04/word. At FPE throughput of 2,000 words/day (~250 words/hour), the same translator at $40/hour matches $0.16/word. The hourly equivalent varies enormously based on actual MT quality.

When hourly works well: - First project with an unknown MT engine - Short projects where per-word setup doesn’t justify the risk assessment - Content domains where MT quality is notoriously unpredictable (legal nuance, creative elements mixed into technical content) - Post-editors who can accurately track and report their time

When hourly breaks down: - Clients resist it. Open-ended costs create approval friction in procurement and budget planning. Most clients want a fixed number before they sign off. - Translators may work slower than necessary when time is the metric (not accusing anyone of bad faith - it’s just a natural dynamic when effort isn’t the variable) - It’s hard to compare across agencies, which makes competitive quoting awkward

The practical fix: always pair hourly rates with a time cap and a word-count estimate. “40,000 words, estimated 12 hours at $50/hour ($600), maximum charge $800.” That gives the client a budget anchor and still protects you if MT quality is worse than expected.

A GTS Translation survey of 212 freelancers in 2025 found translators increasingly pushing for hourly rates or tiered pricing as a response to underestimated post-editing effort. This is a rational reaction: when 66% of translators say MT output requires significant edits and 21.74% say it requires extensive rework, flat per-word discounts create a systematic earnings gap.

Effort-based pricing (TER) - the fair model that’s hard to implement¶

Effort-based pricing is conceptually the most elegant solution: you pay for the actual editing work, not for source words or time. The mechanism is Translation Edit Rate (TER) - tracking what percentage of each segment was changed by the post-editor.

The basic logic from Hunnect’s fair pricing framework: segments modified less than 50% get paid at roughly half price; unchanged segments get paid at a quarter price or less. The more you change, the more you earn. The payment scales with actual effort.

A more sophisticated version: subtract editing distance from 100%, and pay the segment as if it were a fuzzy translation memory match at that percentage. If the post-editor rewrote 20% of a segment, they’re paid as if it were an 80% TM match. This maps onto the TM discount structure that agencies already use, making it familiar territory for both sides.

When effort-based works well: - Long-term relationship between agency and translator where both trust the TER tracking - Large, consistent content domains (product catalogs, software UI strings) where TER reflects actual effort reasonably accurately - When you have CAT tooling that tracks and reports editing distance automatically - If both parties want to eliminate the “who bears MT quality risk” debate permanently

When effort-based breaks down: - TER doesn’t capture cognitive effort. Fixing a subtle mistranslation that looks fluent might require only 2 word changes but 10 minutes of research and expert judgment. TER pays for the 2-word change, not the 10 minutes. - It requires specialized tooling and consistent workflow discipline to track editing distance accurately - Disputes are more likely because both parties need to agree on what counts as an “edit” - Clients can’t easily understand or verify the billing - it requires trust and transparency that takes time to build

As noted in localization workflow analysis, the TER metric creates a perverse incentive when it’s the only variable: it pushes editors toward making stylistic changes for rate reasons rather than quality reasons, and discourages the kind of deep research that good post-editing sometimes requires.

The maximum-charge model: the best hybrid¶

The Hunnect approach proposes what they call the maximum-charge model, which addresses the biggest risk in all three models: the translator getting underpaid because MT quality was worse than expected.

Here’s how it works:

1. Calculate the maximum charge: what you’d bill if every segment needed full retranslation at your standard human rate. This is the ceiling.
2. Track actual editing effort (either by TER or by time).
3. Invoice based on actual effort, not to exceed the maximum.

If editing effort turns out to be light (good MT quality), the client pays less than the ceiling. If editing effort is heavy (poor MT quality), the translator still earns fairly because the billing rises toward the maximum. Everyone knows the worst case upfront.

This eliminates the most common MTPE dispute: a translator doing near-human-translation effort at post-editing rates because MT quality was worse than the agency expected when quoting.

The model works well when paired with a minimum charge - a floor rate for small projects where the overhead of tracking doesn’t justify going below a fixed minimum.

How each model affects each party¶

The pricing model doesn’t just affect total cost - it redistributes risk between agencies, translators, and clients differently.

For translation agencies doing high-volume MTPE work, the maximum-charge hybrid solves the most expensive problem: the translator who pushes back mid-project because the MT output isn’t what was promised.

For freelancers, hourly with a time estimate is the safest starting position with any new client or MT engine - shift to per-word once you’ve validated the quality baseline.

What the translator side actually wants¶

GTS Translation’s 2025 survey of 212 freelancers provides the most direct data on this:

“Around 50% of respondents do not offer discounts for MTPE work, arguing that post-editing can take as much time as traditional translation.”

Among those who do discount: the most common range is 10-30%, not the 40-60% that clients often expect. And 80% of translators believe MTPE has put downward pressure on pricing expectations - a structural problem for everyone if post-editing rates compress to the point where skilled linguists stop doing the work.

This matters for agencies choosing a pricing model: if your model systematically under-compensates post-editors, you’ll have trouble retaining skilled translators for MTPE work. The cheapest per-word rate you can advertise is only sustainable if the translators doing the work can earn a reasonable daily rate.

Picking the right model: a decision framework¶

Use this as a starting point for each new project:

Start with per-word if: - You’ve run at least one pilot project with this MT engine and language pair - Content type is consistent and formulaic (technical docs, product catalog, software UI) - Volume is over 10,000 words (the per-word model scales well at volume) - MT quality pilot showed 80%+ fluency with mainly terminology issues

Use hourly if: - This is the first project with a new MT engine or untested language pair - Content includes mixed material where effort varies segment by segment - Project volume is under 5,000 words (overhead of per-word piloting isn’t worth it) - The post-editor you’re working with specifically requests it

Consider effort-based if: - You have an established long-term relationship with a specific translator - You have CAT tooling that tracks editing distance automatically - Content is consistent enough that TER is a reasonable proxy for effort - Both parties are willing to document the TER framework in writing

Use the maximum-charge hybrid if: - MT quality is genuinely unknown upfront - You want to protect both the translator and the agency from quality surprises - You’re building a long-term MTPE relationship and want a sustainable model from day one

Language pair matters more than model choice¶

One variable that overrides any model decision: MT quality for the specific language pair.

SwissGlobal’s 2026 productivity analysis documented the spread: - English → French: +130% productivity boost with MTPE - English → Polish: +18% boost - English → Swedish: -7% (slower than human translation)

For EN-FR, LPE at $0.06/word on a 50,000-word project ($3,000) might deliver the same output that human translation at $0.20/word ($10,000) would. That’s a 70% saving that any pricing model can capture.

For EN-SV on the same project, MTPE doesn’t make economic sense at any pricing model - the post-editor would take longer than a human translator, so you’d pay MTPE rates for human translation effort. Run a pilot. Always.

Rates also increase 20-50% for language pairs with lower MT performance, like English-Arabic or English-Japanese, reflecting the higher effort required regardless of pricing model.

Tying it to contracts: what to put in writing¶

Whatever model you choose, three things need to be in the contract before work starts:

1. Post-editing level. Explicitly state whether this is light or full post-editing. ISO 18587:2017 - the international standard for MTPE - requires the editing level to be agreed in writing before work starts. If it’s FPE, output must meet the same quality bar as human translation. If it’s LPE, both parties acknowledge it’s not publication-ready.

2. What happens if MT quality is poor. Define the threshold: if post-editing time exceeds X words/hour or editing distance exceeds Y%, the pricing model renegotiates or shifts to full human translation. Without this clause, a quality surprise has no resolution other than a dispute.

3. Pilot clause for new pairs. For any new language pair or MT engine, require a 2,000-5,000 word pilot at agreed rates before committing to scale. This is cheap insurance against the scenario at the start of this article.

Where AI document translation tools fit in¶

For agencies handling high-volume formatted documents - technical manuals in DOCX, reports in PDF, multilingual product datasheets - the friction in MTPE workflows often comes from the file handling layer, not the linguistic layer: stripping formatting, rebuilding translated output, preserving tables through the MT stage.

AI document translation platforms like ChatsControl handle this file-handling step: upload a DOCX or PDF, get back a formatted draft with the original layout preserved, route it to the post-editor. The linguist focuses on the editing; the platform handles the formatting. Worth considering for agencies where file prep and reassembly create bottlenecks on per-word MTPE projects. The post-editor still does the linguistic work - this affects the file-handling part of the workflow, not the choice of pricing model.

FAQ¶

Which MTPE pricing model do most agencies use?¶

Per-source-word is the default - used by approximately 70% of agencies and preferred by 69.6% of translators (Nimdzi survey). It scales with volume and clients understand it immediately for comparison. Hourly and effort-based models are used in specific situations: new MT engines, unknown content domains, or long-term bilateral relationships where TER tracking is already set up.

Is per-word or hourly better for MTPE freelancers?¶

Hourly is safer when MT quality is unpredictable - it protects you if output is poor and you end up retranslating. Per-word pays better when MT is clean and you’re fast. If you’re working with a new agency or untested engine, negotiate hourly with a minimum guarantee, then switch to per-word once you’ve established a baseline for MT quality.

What is TER pricing in MTPE?¶

TER (Translation Edit Rate) pricing links payment to how much you actually change the MT output. Segments you barely touch get a reduced rate; segments you rewrite get full rate. It’s the fairest model in principle, but doesn’t capture cognitive effort - fixing a subtle mistranslation might require only 2 word changes but 10 minutes of research. Works best in established relationships where both parties know the content type.

What is the maximum-charge model for MTPE?¶

A hybrid: the maximum charge is set at full human translation rates (the ceiling if every segment needed retranslation). Actual invoicing is based on editing effort. If you lightly edited 80% and heavily rewrote 20%, you’re paid proportionally. It protects translators from under-compensation and agencies from overpaying when MT quality is good. Recommended by Hunnect as the most sustainable bilateral model.

Why do some translators refuse MTPE discounts?¶

Because the efficiency promise often doesn’t materialize. A GTS Translation survey of 212 freelancers in 2025 found 66% say MT output requires significant edits, and 21.74% say it requires extensive rework - effectively retranslation at a lower rate. When post-editing effort approaches 80-100% of human translation effort, taking a 40-50% rate cut means working well below market hourly earnings.

How do MTPE rates vary by language pair?¶

Significantly. SwissGlobal documented EN-FR post-editing at +130% speed vs human translation, EN-PL at only +18%, and EN-SV at -7% (slower). For difficult pairs, MTPE rates need to approach human translation rates. Always pilot 5,000 words before committing any new language pair to a discount model.

Sources¶

• Nimdzi - MTPE pricing preferences and industry survey
• Hunnect - Fair MTPE pricing and the maximum charge model
• GTS Translation - The State of MTPE in 2025: What Translators Think
• Artlangs - MTPE Pricing Guide 2026
• Weglot - MTPE rates breakdown
• ISO 18587:2017 - Post-editing of machine translation output: requirements
• SwissGlobal - Translation productivity 2026: human vs MTPE