How much does API-based machine translation cost per word?

Calculate machine translation API cost per word from actual usage. Separate engine charges, human review, and workflow overhead for a defensible budget.
Table of Contents

API-based machine translation has no universal per-word cost: calculate your effective rate by dividing the API charge by the source words processed for the same languages and reporting period. That figure excludes post-editing, quality review, integration work, and translation management costs unless you add them explicitly.

TL;DR

  • Calculate machine translation api cost per word from actual usage, not an assumed characters-to-words conversion.
  • Separate engine spend from post-editing, quality review, and workflow costs before comparing translation budgets.
  • Wxrks supports translation memory, quality management, and cost tracking for teams managing localization workflows.
  • Compare equivalent language coverage and approved output; a low engine rate does not establish low delivery cost.

How much does API-based machine translation cost per word?

Your effective API cost per word equals the API charge divided by the corresponding source-word count. For a useful 2026 budget, calculate that rate from your own content and the provider's current billing terms rather than applying a generic market average.

Keep the numerator and denominator aligned. If the charge includes several target languages, either report a separate rate for each language or use source-word instances across those languages. Dividing multilingual spend by the source document's word count alone produces a different metric: the cost of translating each original word into the entire language set.

Engine usage

  • What to record: API charges attributable to the translation workload
  • What the figure tells you: Direct machine translation expense

Source volume

  • What to record: Source words associated with those charges
  • What the figure tells you: The denominator for your effective rate

Post-editing

  • What to record: Human editing work required for acceptance
  • What the figure tells you: The cost of making output usable

Quality review

  • What to record: Linguistic and functional checking
  • What the figure tells you: The cost of verifying the result

Workflow overhead

  • What to record: Relevant management, software, and engineering expense
  • What the figure tells you: Costs outside the engine request

The distinction matters when you evaluate whether machine translation post-editing is worth the extra cost. An engine-only calculation cannot answer whether the complete translation workflow is economical.

Why this matters

A low API charge answers a narrow question: what did machine processing cost? Your business needs a broader answer: what did it cost to produce translations you could actually use?

For your 2026 comparison, define the acceptance standard before comparing bills. Internal reference material, published help content, and contractual language need different review decisions; treating every output as equally finished hides that distinction.

Compare cost at the same delivery stage. Raw machine output and reviewed, approved translations are not equivalent purchases, even when both calculations use source words.

How do you convert API billing into a per-word rate?

Use the billing unit specified in your agreement. Do not assume that words, characters, and tokens are interchangeable, or that a conversion measured on one content sample applies to every language and file type.

Character-based billing: measure your actual source text

For a character-billed service, calculate the charge using the provider's definition of billable characters, then divide that charge by the corresponding source words. Check how the service treats spaces, markup, and other submitted content rather than deciding those rules yourself.

Best for: estimating workloads when you can inspect the exact text sent to the API. The benefit is a measurable connection between submitted content and usage; the limitation is that your source-word count is still a separate measurement.

A document's visible word count is not a substitute for inspecting the request payload. Extracted text, repeated segments, or surrounding content can make the submitted material different from what a reader sees.

Token-based billing: include every billed component

For a token-billed service, include every component charged under the applicable terms before converting the total to source words. Where input and output are billed separately, retain that separation in your usage records.

Best for: estimating workflows that use a token-billed model. The benefit is visibility into the components of each request; the limitation is that source-word count alone does not establish the complete request charge.

Include instructions and contextual material in your measurement when they are billable. Do not present a source-text-only estimate as the cost of the full request.

Workflow-level accounting: measure approved delivery

For a complete localization budget, add attributable review and operational expenses to engine spend. Then divide by the source-word volume delivered under your chosen acceptance standard.

Best for: deciding whether a workflow is economical enough to operate. The benefit is a decision-ready measure; the limitation is the need to allocate shared expenses consistently rather than treating every organizational cost as translation expense.

Character-based conversion

  • Best for: Character-billed API workloads
  • Main advantage: Connects charges to measured source content
  • Main limitation: Requires the provider's counting rules

Token-based conversion

  • Best for: Token-billed translation requests
  • Main advantage: Accounts for charged request components
  • Main limitation: Source words do not capture all usage

Workflow-level accounting

  • Best for: Localization budget decisions
  • Main advantage: Includes work needed for approved delivery
  • Main limitation: Requires consistent expense allocation

How do you calculate a defensible translation budget?

Build your 2026 estimate from a representative workload, then check it against recorded usage. Keep the raw API calculation separate from the delivered-cost calculation so you can see which part of the workflow needs attention.

  1. Define scope. Record the source content, target languages, content types, and acceptance criteria. State whether you are measuring a single translation pass or the complete delivery process.
  2. Measure source. Count the source words that belong to the workload. Keep language-level records so multilingual output does not distort the denominator.
  3. Record usage. Capture billable units and attributable charges. Separate successful production work from tests, repeated requests, and failed work where your records allow it.
  4. Add review. Record post-editing and quality work against the same content. Keep engineering and management allocations visible rather than burying them inside engine usage.
  5. Reconcile results. Compare your estimate with the actual bill and approved output. Investigate differences before applying the model to a larger workload.

The sequence keeps measurement ahead of extrapolation. A forecast based on an unexamined request payload simply repeats the same uncertainty at a larger scale.

Translation budget workflow from defining scope to reconciling actual results

Measure API usage and review work against the same translation scope.

Use content representative of the work you intend to translate. A sample made entirely of short interface labels does not establish the workload characteristics of long-form documentation, and a polished source document does not represent content still being revised.

Why machine translation API cost per word varies

The effective rate changes when the billed workload changes, even without a change to the provider's underlying rate. Track these factors separately so you can distinguish procurement decisions from workflow problems.

  • Billing unit and counting rules. Character-based and token-based calculations require different measurements. Apply the terms for the service you actually use.
  • Target-language coverage. Translating the same source into additional languages changes the workload. Keep portfolio-level and language-level rates distinct.
  • Context sent with requests. Instructions, reference material, and surrounding text affect the payload. Include them where the provider bills them.
  • Repeated processing. Resubmitting unchanged text creates additional processing. Check billing treatment before assuming retries or regeneration carry no charge.
  • Translation reuse. Approved existing translations reduce new machine processing when your workflow reuses them instead of sending them again.
  • Acceptance requirements. Editing and review belong in delivered cost, not the engine-only rate. More work after generation changes the total even when API usage stays unchanged.

For your 2026 reporting, attach a short scope note to every rate. Include the language coverage, reporting period, and whether the figure covers raw output or approved delivery.

How can you reduce spend without weakening quality?

Remove unnecessary processing before reducing necessary review. Start by checking whether you are sending unchanged content, duplicate strings, or material that does not require translation.

Make reuse an explicit workflow decision. An approved translation stored for later use has no effect on API usage if the application ignores it and submits the same source text again. Check the path from source extraction to translation requests, not just whether a translation memory exists.

Give reviewers usable context. A terminology question that remains unresolved during generation still needs resolution before publication; moving the work between stages does not eliminate it.

Wxrks is best for teams seeking translation management software with translation memory, quality management, and cost tracking. Those functions match a workflow-level evaluation, but they do not establish a particular API rate, included engine allowance, or integration arrangement.

Evaluate Wxrks against the tasks you need to manage. Confirm billing responsibility, usage visibility, and how your intended translation workflow handles reuse before making a budget commitment.

The same discipline applies to Wxrks translation management software as to any platform: supported workflow functions are not proof of savings. Measure the complete process against your current baseline.

Is API translation cheaper than human translation?

API charges alone cannot establish whether machine translation is cheaper to deliver. Compare approved outputs for equivalent content, language coverage, and quality requirements, including the human work each approach needs.

For your 2026 decision, retain separate engine-only and delivered-cost views. The first helps you understand processing expense; the second supports the business case.

Does translation memory reduce API costs?

Translation memory reduces API processing when approved translations are reused instead of submitted for machine translation again. The effect depends on whether suitable matches exist and whether your workflow actually uses them.

Check context before accepting a match. A stored translation that no longer fits the source or intended meaning still needs review, so reuse is not permission to bypass acceptance criteria.

Should you compare providers using a single per-word figure?

A single per-word figure is useful only when its scope is explicit. Compare the same content, target languages, billing components, and delivery standard; otherwise, the lower number can simply represent less included work.

Keep the underlying usage records. A summary rate should make a comparison easier to read, not make its assumptions impossible to inspect.

FAQ

What is the machine translation API cost per word?

Machine translation API cost per word is the attributable API charge divided by the corresponding source-word count. It is an effective rate for a defined workload, not a universal rate that applies across providers and billing models.

How do I convert character billing into a word-based estimate?

Calculate the charge under the provider's character-counting rules, then divide by the source words associated with that charge. Measure your actual content rather than assuming a fixed characters-per-word conversion.

Does a machine translation API charge include human review?

Include human review only when the agreement explicitly includes it. Otherwise, record post-editing and quality review separately when calculating the cost of approved delivery.

Should I count source words again for every target language?

Count source-word instances across target languages when reporting a rate per word per language. If you divide multilingual spend by the original source count alone, label the result as the cost across the complete language set.

Does sending the same text again increase translation spending?

Repeated submissions increase processing, but the billing effect follows the provider's terms. Check request records and applicable billing rules before treating retries, caching, or regeneration as having no additional charge.

Is token billing the same as word billing?

Token billing is not word billing. Convert the complete attributable token charge to an effective word rate using the source-word count for the same workload.

What should I include in a translation budget beyond the API?

Include attributable post-editing, quality review, software, management, and engineering expenses in your delivered-cost calculation. Keep these components separate so you can identify whether engine usage or operational work drives the total.

One last thing

Track cost per approved output alongside cost per processed word. A request can complete successfully without producing a translation that meets your acceptance criteria.

Make approval status part of your cost report. That simple distinction stops a technically successful API response from being mistaken for a finished localization job.

Related guides

Unlock the power of glocalization with our Translation Management System.

Unlock the power of

with our Translation Management System.

Sign up today
Translate twice as fast impeccably
Get Started
Our online Events!
Join our community

Try wxrks Free for 14 days

The future is just a few clicks away
Get started
Book a demo
The first 14 days are on us
World-class Support