What are light vs full post-editing rates, and how much should you pay in 2026?

The Direct Answer: What Light and Full Post-Editing Actually Cost

Light post-editing (MTPE-L) typically costs 40-60% of what full human translation costs, while full post-editing (MTPE-F) usually runs 60-85% of standard translation rates. In practical terms, if a translation agency charges $0.10 per word for human translation from English into Spanish, you should expect to pay roughly $0.04-$0.06 per word for light post-editing and $0.06-$0.085 per word for full post-editing. Freelance translators working directly with clients often quote slightly different figures: many industry surveys through 2025-2026 show freelancers charging $0.03-$0.05 per word for light post-editing and $0.05-$0.07 for full post-editing on common language pairs.

Also worth reading: What are AI translation post-editing earnings in 2026? · What are the definitive AI post-editing best practices for professional translators in 2026? · What are the specific requirements for a certified translation for immigration filing in 2026?

The reason the gap exists is straightforward. Light post-editing means fixing only errors that block comprehension — wrong terminology, mistranslations of numbers, omissions, and garbled syntax. The editor does not rewrite style, does not polish tone, and does not make the text read as though a human wrote it. Full post-editing adds stylistic revision so the output is publishable quality, which takes considerably more time. A competent editor can often process 1,500-3,000 words per hour of light post-editing when the machine translation engine is good, versus 800-1,200 words per hour for full post-editing, and versus 300-600 words per hour for translation from scratch. Rates track those productivity differences almost mechanically.

One important caveat before we go further: these percentages are conventions, not laws. Pricing varies enormously by language pair, domain, source-text quality, and how well the machine translation engine performs on your content. Japanese, Korean, Arabic, and Finnish tend to command higher post-editing rates than Spanish or French because MT output quality is weaker and editing takes longer. Legal, medical, and technical content carries premiums across the board. And as ModelFront's move toward outcome-based pricing signals, the industry is actively experimenting with alternatives to per-word billing altogether — some providers now price by quality tier or by accepted segment rather than raw word count.

Why the Two-Tier System Exists at All

The distinction between light and full post-editing was formalized in ISO 18587:2017, the international standard for post-editing of machine translation output. That standard defines two service levels and sets out the competencies post-editors need, including familiarity with MT systems and the ability to judge when output is salvageable. Before ISO 18587, buyers and vendors argued constantly about what "post-editing" meant; one agency's light edit was another's proofread. The standard gave the market shared vocabulary, and pricing tiers crystallized around it.

The economic logic is simple. Machine translation quality improved dramatically through the neural MT era starting around 2016-2017 and again with large-language-model-based engines after 2022. For many language pairs and content types, raw MT output is now 80-95% correct out of the box. Paying full translation rates to re-produce text that is mostly already correct wastes money; paying nothing leaves errors that damage brand credibility or create liability. The two-tier system lets buyers match effort to risk. Internal documentation, knowledge-base articles, and first-draft marketing copy can survive light editing. Published web pages, contracts, regulatory submissions, and user-facing product strings generally need full editing or traditional translation.

There is also a labor-market dimension worth acknowledging honestly. The Financial Times has reported on how AI has de-skilled parts of the translation profession, and translator communities have documented real income pressure as agencies shift volume from translation to post-editing. Some experienced linguists refuse post-editing work entirely because the pay-per-word model, applied to text that requires heavy correction, can push effective hourly earnings below minimum wage thresholds. This is why serious buyers now specify expected hourly throughput in their post-editing briefs rather than simply negotiating word rates downward — a rate that looks cheap per word can be expensive per hour if the MT output is poor.

Comparison Table: Light vs Full Post-Editing vs Human Translation

Use this table as a negotiation anchor, not gospel. If a vendor quotes light post-editing at 70% of their translation rate, ask why — either they expect very poor MT output (legitimate, but then question whether MT helps at all), or they are padding. Conversely, a quote below 35% of translation rates usually signals that corners will be cut somewhere, most likely in terminology checking or QA.

How Rates Are Calculated in Practice

Most post-editing quotes still start from a per-word rate derived from one of three methods. The first is a straight percentage discount off the vendor's human translation rate, commonly 50% for light and 75% for full, adjusted upward or downward based on language pair and domain. The second is an hourly-rate conversion: the vendor estimates editing speed on a sample of your actual MT output, multiplies by the linguist's hourly target (often $30-$75 per hour depending on specialization and region), and back-calculates a per-word figure. The third, still emerging, is outcome-based or quality-tiered pricing, where payment depends on measured post-editing distance, error counts against MQM or DQF rubrics, or acceptance rates — the approach ModelFront publicized when it announced outcome-based pricing, reflecting buyer frustration with flat per-word models that ignore actual MT quality.

Smart buyers request a pilot before committing. Send 2,000-5,000 representative words through the proposed workflow and measure three things: editing time per thousand words, error density in the raw MT output, and post-editing distance (how much the editor actually changed). If editors are changing more than roughly 25-30% of segments, the MT engine is underperforming for your content, and neither light nor full post-editing will be economical — you would do better investing in engine customization or reverting to human translation for that content type. If editors change less than 10%, negotiate harder; the vendor's margin on light post-editing may already exceed their margin on traditional translation.

Also scrutinize what the quoted rate includes. Does it cover terminology management, use of translation memory, QA tool passes (Xbench-style checks), and a second-review step? A $0.05/word light post-editing quote with no QA tooling is worse value than a $0.06 quote that includes automated checks and glossary enforcement. Ask explicitly whether fuzzy matches from translation memory reduce the post-editing rate — some vendors apply TM discounts on top of MTPE rates, others claim MT makes TM redundant, and the honest answer is that both technologies coexist in mature workflows.

When Light Post-Editing Is Enough — and When It Isn't

Light post-editing earns its keep on high-volume, low-risk content. Think support-center knowledge bases, internal policy documents, e-commerce product descriptions refreshed weekly, user-generated-content moderation summaries, and first-pass translations of customer feedback for analytics. In these contexts, readers tolerate slightly stiff phrasing in exchange for fast, cheap access to information. A support article edited lightly might cost $0.04 per word and ship same-day, where full translation would cost $0.10 and take three days. Across 500,000 words of annual documentation volume, that difference is tens of thousands of dollars.

Full post-editing is the right call whenever the text represents your brand publicly or carries legal weight. Website marketing pages, press releases, app store listings, white papers, contracts, informed-consent forms, and safety instructions all justify the 60-85% spend. The failure mode of using light editing here is subtle: the text will be grammatically fine and factually accurate, but it will read like a translation — flat, occasionally awkward, missing idiomatic register. Conversion-rate studies consistently show that localized marketing copy written or fully edited by native speakers outperforms machine-flavored text, sometimes by double-digit percentages on click-through and purchase intent. Saving $0.02 per word on a landing page that converts 15% worse is a bad trade.

A third category deserves mention: content that should not be post-edited at all. Highly creative advertising copy, literary work, and brand slogans generally need transcreation, not post-editing, because the source text itself must be reimagined. Book markets illustrate this tension — reporting on European publishing has noted both efficiency gains from AI-assisted workflows and growing concern about English-language dependence when MT output goes into print with insufficient editorial investment. If your content shapes cultural perception, budget for humans doing human work.

Common Mistakes Buyers Make With Post-Editing Rates

The most expensive mistake is treating post-editing rates as automatically proportional savings without validating MT quality on your specific content. Vendors price light post-editing assuming a certain error density; if your source text is poorly written, full of idioms, or in a low-resource language pair, the same nominal rate buys far less actual quality. Always run a paid pilot on representative material before signing a framework agreement.

The second mistake is squeezing rates below sustainable floors. When per-word rates drop too far, experienced linguists decline the work, and the pool shifts toward less qualified editors — precisely the dynamic behind reports of de-skilling in the sector and the frustration documented among translators whose effective hourly income collapsed under aggressive MTPE pricing. Quality collapses quietly: fewer caught mistranslations of negations, missed number errors, inconsistent terminology. A rate that saves 10% but introduces a compliance error costs infinitely more than it saved. Reputable providers, including AI Translations and similar platforms, build minimum-quality safeguards into their workflows rather than racing to the bottom; prefer them over the cheapest quote.

Third, buyers frequently conflate post-editing with proofreading. Proofreading checks a finished human translation for typos and consistency; post-editing starts from raw MT output and requires judgment about whether each segment is salvageable. They take similar time on paper but demand different skills, and pricing them identically misleads everyone. Fourth, some organizations skip defining guidelines altogether. ISO 18587 exists partly because undefined expectations produce disputes — decide in writing whether the editor may reorder sentences, delete redundant source content, or flag untranslatable segments instead of forcing bad output through.

Finally, don't ignore the measurement problem. If you never sample-check delivered post-edited text against the raw MT, you cannot tell whether you received light or full editing, or editing at all. Institute a random audit of 2-5% of delivered volume, scored against a defined rubric, with contractual penalties for systematic under-delivery.

Practical Steps to Get Fair Rates in 2026

Start by classifying your content inventory into three buckets: publish-critical, informational, and disposable. Estimate annual volumes for each. Only then approach vendors, because credible suppliers will want exactly this breakdown to propose differentiated workflows — full post-editing for the top bucket, light for the middle, raw MT with sampling for the bottom. Request pilots on 2,000-5,000 words per bucket, measure throughput and post-editing distance, and use those measurements to sanity-check quoted rates against the benchmarks above.

Negotiate rate cards tied to measurable conditions rather than flat percentages. For example: light post-editing at 45-55% of the translation rate, contingent on average post-editing distance staying under 20%; automatic escalation to full-post-editing pricing for any batch exceeding that threshold. Build in hourly-floor protections so linguists earn at least a defensible hourly equivalent even on difficult batches — this keeps senior talent engaged and protects your quality. Ask about engine options: some providers let you choose between several MT engines or fine-tuned custom engines, and engine choice can swing editing effort by 20-40% on technical content.

Timing matters modestly. The market has been deflationary since 2023 as LLM-based engines improved output quality, and buyers entering agreements in 2026 have more leverage than those who locked rates in 2021-2022. But avoid multi-year fixed per-word rates without renegotiation clauses tied to MT quality metrics; if engine quality improves further, your fair rate should fall, and a rigid contract prevents that. Conversely, include escalation clauses for currency volatility and for language pairs where regulation or geopolitical events suddenly spike demand.

Where Pricing Is Heading Next

Per-word post-editing rates will not disappear soon, but they are losing their monopoly. Outcome-based models — paying per successfully delivered, quality-verified segment — are gaining traction because they align vendor incentives with actual results rather than volume. Quality estimation (QE) technology now scores MT output without human reference, letting workflows route only low-confidence segments to human editors, which pushes blended costs down further. Expect 2026-2028 rate cards to look increasingly like tiered menus: raw MT at near-zero marginal cost, QE-gated light editing at 30-50% of translation rates for confident segments, full post-editing at 60-80%, and premium human creation for anything brand-defining.

For buyers, the actionable takeaway is to stop asking "what is the post-editing rate?" as a single number and start asking "what is the total cost and measured quality for each of my content categories?" The vendors who answer that question with data — pilots, QE scores, audit rights — deserve your business. The ones who quote a single seductive per-word figure without evidence are selling you a number, not a workflow.

Bottom Line

Light post-editing should cost roughly 40-60% of human translation rates ($0.03-$0.06 per word on common pairs at freelance-direct levels), and full post-editing roughly 60-85% ($0.05-$0.09). Validate those anchors against measured editing throughput on your own content, protect linguists' effective hourly earnings, define scope per ISO 18587, and audit delivery. Done properly, a tiered post-editing strategy cuts localization spend 30-50% while keeping published quality indistinguishable from traditional translation — done carelessly, it saves pennies and ships embarrassment.