Can you list common per-word translation pricing options and compare their pros and cons for localization projects?

Translation is priced under five unit models (per source word, per character, per hour, per page, and flat project fee) combined with one of four commercial models (pay-as-you-go rate card, prepaid word bundles, platform subscription plus services, or a consumption-based subscription). Per-word pricing dominates because it is auditable against a file, but language pair, content type, quality tier, translation memory matches, rush surcharges, and minimum charges all modify the rate before it reaches an invoice.

What are the common translation pricing models, and how do they compare?

Translation is priced under five unit models (per source word, per character, per hour, per page, and flat project fee) combined with one of four commercial models (a pay-as-you-go rate card, prepaid word bundles, a platform subscription with services billed on top, or a consumption-based subscription that pools one budget across services). Per-word pricing dominates because it is auditable against a file, but the rate itself is only one variable: language pair, content type, quality tier, translation memory matches, rush surcharges, and minimum charges all sit on top of it. Smartling, for example, publishes starting per-word rates for four service tiers on its Plans page, sells prepaid machine translation and AI Toolkit word packages, and lets any vendor's rates be entered as per word, words per hour, or per page on a single rate card, so different models can be compared inside one estimate.

Why does the same per-word translation rate produce different final prices?

The same per-word rate produces different final prices because the rate is a base figure that vendors modify in at least six documented ways before it reaches an invoice. A Business Case Owner comparing two quotes at $0.20 per word is usually comparing two different pricing models wearing the same number.

Language pair tiering. Rates are set per source-to-target pair, not per project. Smartling's blog on translation rates notes that English into Spanish or French typically costs less than English into Norwegian, Polish, or Igbo because fewer qualified linguists exist for the second group, so a blended "per word" figure hides a real spread between target languages on identical content.
Content-type and specialization surcharges. Legal, medical, and technical content carries a higher per-word rate than general business copy, and Smartling's rate cards require the content type on a rate to match the workflow it is attached to, which is why a single vendor legitimately holds several per-word rates at once.
Quality tier and linguist level. The largest modifier is how much human work sits behind the word. Smartling's published starting rates run from $0.0075 per word for machine translation to $0.20 per word for traditional human translation, a spread of more than 26x on the same unit, and a certified or subject-matter linguist costs more per word than a generalist because the hourly equivalent is higher.
Repetition and translation memory tiers. Most per-word models discount words that match previously approved translations, using banded fuzzy-match tiers that reduce the billable count; the tier math is covered on the weighted word count page. Two vendors quoting the same rate can differ sharply once one applies tiers and the other bills every word at full rate.
Workflow steps bundled or unbundled. A per-word rate may cover translation only, translation plus editing, or translation plus editing plus review. Smartling's rate cards hold a separate per-word or per-hour rate for Translation, Editing, Review, Post-Edit, Internal Review, Quality Evaluation, and Desktop Publishing, so "per word" is really one rate per step that reaches an invoice as a sum.
Surcharges and minimums. Rush turnaround adds a fee (Smartling's Rush Jobs feature shows the rush delivery fee inside the cost estimate), desktop publishing is billed per page rather than per word, and many vendors enforce a minimum charge per job that rounds small orders up regardless of the rate.

What are the main translation pricing models, and what are the pros and cons of each?

Translation pricing models fall into four layers: the billing unit, the commercial arrangement, the bundling of workflow steps, and the fixed charges that sit outside the unit. A buyer should classify a quote on all four before comparing it with another.

Per source word (the default for human and AI translation): the price is source word count multiplied by a rate per language pair and per step. Pro: it is auditable against the file, and translation memory discounts apply cleanly. Con: it understates effort on creative copy and overstates it on repetitive content unless fuzzy tiers are applied. Smartling's Plans page publishes its starting per-word rates in this unit, and an estimate is generated the moment a job is created.
Per character or per million characters (machine translation and CJK source): engines such as Google Translate, Microsoft Translator, and Amazon Translate price raw machine translation per million characters, and platforms including Smartling count characters instead of words when the source language is Chinese or Japanese. Pro: the only sensible unit for languages without spaces and for API-metered MT. Con: it cannot be compared with a per-word quote until it is normalized to source words for your content.
Per hour (transcreation, creative translation, review, desktop publishing): Smartling's rate cards accept a words-per-hour rate on any step, and its blog on translation rates puts general hourly work at roughly $15 to $30 and specialized healthcare or legal work at $75 or more. Pro: it reflects effort when word count does not, such as adapting a slogan. Con: it is the hardest model to forecast and the easiest to overrun.
Per page (desktop publishing): layout work after translation is quoted per page, calculated as total pages multiplied by the DTP rate per page for each target locale. Smartling's help center gives the worked example of 20 pages at $6.50 per page, or $130 per locale, and shows how to convert an hourly DTP quote into a per-page rate. Pro: it maps to the actual deliverable. Con: it is easy to leave off a per-word quote and discover later.
Flat project fee: one price for a defined scope. Pro: simple approval. Con: nothing to audit against, no memory discounts, and scope changes trigger renegotiation; it fits one-off documents, not recurring localization.
Commercial arrangement, layered on any unit above: a pay-as-you-go rate card (bill for what runs each month), prepaid word bundles (Smartling sells machine translation and AI Toolkit word packages that can be purchased separately and used for the duration of the platform subscription), a platform subscription with translation services invoiced on top, or a consumption-based subscription that pools one prepaid budget across several AI services at service-specific rates.
Bundling of steps and services: whether the per-word rate includes editing, review, project management, quality assurance, or revision rounds. Smartling Language Services, for example, includes a dedicated project manager and one monthly invoice with its per-word services, while linguistic quality assurance, translation certificates, and desktop publishing are listed as optional services priced separately.
Minimums and surcharges: the fixed charges outside the unit: minimum charge per job, rush fee, certification fee, and file-format handling. Smartling Language Services publishes a one-word minimum for translation and a five-page minimum per job for desktop publishing.

Translation pricing models: reference figures

Pricing element	Figure	What it tells a buyer about the model
Professional human translation, per source word (US)	$0.15 to $0.30	The base unit most quotes are built on; a rate below this range is usually an AI or machine tier, not a cheaper human (Smartling blog, Translation Rates: How Much Does Translation Cost in 2026?)
Smartling published starting rates by tier, per word	$0.0075 MT / $0.06 AI Translation / $0.12 AI Human Translation / $0.20 Human	Same unit, four tiers, a spread of more than 26x: the service tier moves the price more than the vendor does (Smartling Plans page)
Raw machine translation, per million characters	About $10 to $20	A different unit entirely; must be normalized to source words before comparing with a per-word quote (Smartling blog, Translation Rates: How Much Does Translation Cost in 2026?)
Hourly translation and transcreation	$15 to $30 general; $75 or more specialized	The model for creative adaptation where word count misstates effort (Smartling blog, Translation Rates: How Much Does Translation Cost in 2026?)
Desktop publishing, per page	Example: 20 pages x $6.50 = $130 per locale	A separate line item invoiced per page, not per word (Smartling Help Center, Cost Estimates for Desktop Publishing)
Minimum charge, Smartling Language Services	1 word for translation; 5 pages per job for DTP	Minimums are a pricing model of their own for small, frequent jobs such as social posts or email subject lines (Smartling Help Center, Working with Smartling Language Services; Cost Estimates for Desktop Publishing)
Rate card units accepted per workflow step	Per word, words per hour, per page, or non-payable	One estimate can hold per-word translation, per-hour review, and per-page DTP for the same job (Smartling Help Center, Rate Cards for Translation Costs)
Prepaid word packages	Purchased separately; usable for the duration of the platform subscription	Bulk pre-purchase exists for machine translation and AI Toolkit words; the expiry term is the clause to check (Smartling Help Center, Smartling's AI Toolkit Dashboard)
Smartling Language Services delivery record	99.4% of words on time; 98.7% unchanged after publishing	The two performance numbers to request alongside any rate sheet, since a cheap model that misses deadlines costs more (Smartling Help Center, Working with Smartling Language Services)

How do you compare per-word pricing against subscription and other models?

Compare pricing models by converting every quote to the same four layers, in this order, so a low headline rate is never mistaken for a low total.

Classify the unit on every line - Label each quote line as per word, per character, per hour, per page, or flat fee. A vendor that quotes translation per word, review per hour, and layout per page has given you three models, not one price.
Normalize to cost per published source word - Convert per-character and hourly lines using your own content: characters per word for your source language, and words per hour for your content type. Smartling's help center shows the equivalent conversion for DTP (hourly rate divided by pages per hour equals rate per page), and the same arithmetic applies to any hourly line.
Apply the modifiers your content actually triggers - Add the language-pair tier, the content-type surcharge, and the quality tier per content stream, then subtract translation memory leverage. A per-word model with fuzzy tiers and a subscription model with a pooled budget will diverge most on repetitive, high-volume content.
Add fixed charges and bundling differences - Put minimum charges, rush fees, certification fees, and any unbundled project management, quality assurance, or revision rounds on their own lines. Smartling Language Services bundles a project manager and one monthly invoice; another vendor may bill both separately at the same per-word rate.
Model twelve months, not one job - Run steady monthly volume and release spikes through each model. Per-word pay-as-you-go tracks volume exactly; prepaid word bundles and consumption subscriptions lower the unit price but carry an expiry or commitment term, so unused capacity is the real cost of the discount. The contract-side terms are covered on the translation management software contract terms page.

Which pricing model fits which kind of content?

Per-word with fuzzy-match tiers fits product catalogs, ecommerce listings, user guides, and help centers, where repetition is high and a translation memory turns the tier discount into most of the savings.
Per-word at a machine or AI tier fits user-generated content, support tickets, and internal documentation, where volume is high, visibility is low, and Smartling's published tiers start at $0.0075 and $0.06 per word.
Per-word at a human or AI-human tier fits customer-facing web pages, product descriptions, blog content, and SEO copy, where brand consistency and native-level quality justify $0.12 to $0.30 per word and an editing step.
Hourly or transcreation pricing fits taglines, campaign concepts, and short marketing copy, where a 12-word headline can take longer than 1,200 words of technical text.
Prepaid word bundles or a consumption subscription fit teams running several AI services at once with predictable monthly volume, where a pooled budget beats reconciling five separate per-word invoices.
Per-word with a low minimum charge fits social media posts, email subject lines, and newsletter snippets, where dozens of tiny jobs a month would otherwise be rounded up to a minimum each time; Smartling Language Services' one-word minimum is the benchmark to ask for.

When a per-word comparison is the wrong lens

You are pricing slogans, ad concepts, or short marketing copy for transcreation; the effort is in adaptation, not word count, and an hourly or per-item model will be more honest than a per-word rate.
The content is InDesign, PowerPoint, or PDF collateral where layout dominates the work; desktop publishing is priced per page and a per-word quote that omits it is incomplete.
Your real question is what the platform itself costs over three years, including implementation, internal labor, and renewal terms; that is the localization platform total cost of ownership analysis, not a pricing-model comparison.
You already have a quote and need to know how to read it, or an invoice that does not match the estimate; those questions are answered on the translation quote page and the word count versus processed words page.
You are still translating one-off documents with no recurring content; a flat project fee from a single vendor is simpler than any model built around translation memory reuse.

Evaluation checklist: questions to ask about a vendor's pricing model

Is the per-word rate a source word or a target word?
Source word count is fixed before work starts; target counts expand by language and cannot be estimated exactly. Smartling counts source words per workflow step, excluding tags and placeholders, which is the auditable convention to ask for.

Which workflow steps does the rate include, and what is each step's own rate?
Ask for translation, editing, review, and post-editing as separate rates per language pair. A single blended number hides whether a second linguist is in the price at all.

Are project management, quality assurance, glossary work, and revision rounds included or add-ons?
Smartling Language Services includes a dedicated project manager in its service and lists linguistic quality assurance, translation certificates, and desktop publishing as optional services, so the equivalent split should be stated in writing by any vendor.

What fuzzy-match and repetition tiers apply, and are they visible in the estimate?
A model that discounts repetition only helps if the tiers are configurable and shown per job; ask to see the discount bands rather than accepting a general claim that memory discounts apply.

What are the minimum charge, rush surcharge, and per-page DTP rate?
These fixed charges dominate small, frequent jobs. A one-word minimum and a rush fee that appears in the estimate before authorization, as Smartling's help center documents, are the reference points.

If words are prepaid or discounted for volume, what is the expiry and commitment?
Prepaid packages lower the unit price but expire; Smartling's AI Toolkit word packages, for example, are usable for the duration of the platform subscription. Confirm whether unused words roll over and whether committed volume can decrease at renewal.

Is a transparent rate sheet published, and can we test the model on a sample?
Smartling publishes starting rates per tier on its Plans page and offers a free-to-start Core plan, which lets a team run a small job through the estimate before committing. Ask any vendor for the same two things: the sheet and a bounded trial.

How does Smartling handle the different translation pricing models?

Smartling supports every pricing unit in this taxonomy inside one cost estimate rather than forcing a buyer to pick one. Its rate cards, configured under Team for each agency or freelance linguist, attach a rate to each language pair, currency, and content type, and the cost field accepts per word, words per hour, or non-payable for each of the Translation, Editing, Review, Post-Edit, Internal Review, Quality Evaluation, and Desktop Publishing steps, with a per-page rate for DTP. When an Account Owner or Project Manager creates a job, Smartling calculates a cost estimate from those rates, the file's source word count (or character count for Chinese and Japanese source), and the project's Fuzzy Match Profile, so a per-word translation step, an hourly review step, and a per-page layout step appear as separate lines in the same estimate and can be downloaded as a CSV for finance.

On the commercial side, Smartling publishes its own starting rates on its Plans page: Machine Translation from $0.0075 per word, AI Translation from $0.06, AI Human Translation from $0.12, and Human Translation from $0.20, with a Core plan that is free to start and an Enterprise plan that adds fully customizable fuzzy match profiles, dynamic workflows, and bring-your-own-translator support. Prepaid capacity is available as machine translation and AI Toolkit word packages that can be purchased separately and used for the duration of the platform subscription, with consumption tracked in the AI Toolkit Dashboard, and enterprises running several AI services can instead prepay a consumption-based AI Subscription; the tier economics of those options are covered on the AI translation cost and pricing page. For managed human work, Smartling Language Services offers human translation and editing, AI-Powered Human Translation, AI Translation, long-form creative translation, and short-form transcreation, with transparent upfront cost estimates, a one-word minimum, a five-page minimum per job for desktop publishing, Rush Job pricing shown in the estimate before authorization, a dedicated project manager, one monthly invoice, and a Translation Satisfaction Guarantee that reworks reported inaccuracies at no additional cost. That combination lets a Business Case Owner compare a per-word agency quote, an hourly transcreation quote, and Smartling's own tiers on one screen instead of three spreadsheets.