Direct answer: public machine-translation prices in 2026 are roughly: AWS Translate standard text at USD 15 per million characters, Google Cloud Translation NMT at USD 20 per million characters (first 500,000 characters free each month), Google custom models at USD 80 per million characters, and Google translation LLM at USD 10 per million input characters + USD 10 per million output characters. DeepL Pro Team is EUR 24.99 per user/month with 1 million characters included. Multiple target languages multiply the bill: 1 million source characters into three languages is usually billed as 3 million characters. The API is often the smallest line; terminology, review, layout, and rework decide the budget.
This guide is for marketing, export, localization, product, IT, and procurement teams working on multilingual websites, brochures, manuals, email, and support content. Data cut-off: 31 October 2026. Prices come from official AWS, Google Cloud, and DeepL pages. Azure Translator’s public page shows 2 million free characters/month on F0, but its S1 unit price is dynamically rendered and no stable number was captured, so it is excluded from the numeric comparison. All prices exclude tax; EUR and USD are not converted and no ranking is implied. This is public-source research and an editorial model, not a quote, and it does not promise translation quality, savings, or a go-live date.
APIs bill by source characters, and multiple target languages multiply the count. DeepL bills by user plus included characters, so it is not directly comparable with a per-million price. Public terms, certifications, units, and legal statements need human review; review often costs more than the API.
What are the official translation billing units in 2026?
AWS and Google bill per million characters; Google custom/LLM models add model and input/output differences; DeepL bills per user subscription with included characters. Align the unit first.
There are at least four ways to buy translation: standard NMT API, batch translation per target language, document translation (per page or character), and a subscription-based translation workspace. Prices and units differ; a single “per million characters” number is not a full comparison.
| Service / plan | Billing unit | Official list price | Note |
|---|---|---|---|
| AWS Translate standard text | Million characters | USD 15/million characters | 2 million characters/month free for 12 months after signup |
| AWS Translate batch documents | Million characters | USD 15/million characters | TXT/HTML/DOCX/PPTX/XLSX/XLIFF |
| AWS Translate real-time DOCX | Million characters | USD 30/million characters | Real-time document translation; text/HTML is USD 15 |
| AWS Active Custom Translation | Million characters | USD 60/million characters | Custom output with parallel data; 500k characters/month free for 2 months |
| Google Cloud Translation NMT | Million characters | USD 20/million characters | First 500,000 characters/month free (USD 10 credit) |
| Google custom model | Million characters | USD 80/million characters | Training USD 45/hour, capped at USD 300 per job |
| Google translation LLM | Million input/output characters | USD 10/million input + USD 10/million output | Metered separately; about USD 20/million at 1:1 |
| Google document translation (NMT) | Page | USD 0.08/page | DOCX/PPT/PDF; custom model USD 0.25/page |
| DeepL Pro Individual | User/month + characters | EUR 7.49/user/month (annual), 300k characters | VAT included; not an API unit price |
| DeepL Pro Team | User/month + characters | EUR 24.99/user/month (annual), 1M characters/user | Team admin, SSO (50+ users), glossary |
| DeepL Pro Business | User/month | EUR 49.99/user/month (annual) | “Unlimited” characters under fair use, translation memory, SCIM |
Official list prices in USD/EUR, excluding tax, as of 31 October 2026. AWS/Google are pay-as-you-go list prices; DeepL is a Pro subscription price (annual), not a DeepL API per-character price. A public DeepL API-specific price was not obtained, so only the subscription allowance is used as a comparison; do not treat it as the API unit price. Azure Translator F0 includes 2 million free characters/month, but S1 values were dynamic and not captured.
What do 1M, 5M, and 20M characters cost per month?
Standard NMT is about USD 15–20 per million characters; custom models about USD 60–80; DeepL Team is about EUR 24.99 per million characters at full allowance use.
| Option | 1M characters | 5M characters | 20M characters | Currency |
|---|---|---|---|---|
| AWS standard text | 15 | 75 | 300 | USD |
| AWS real-time DOCX | 30 | 150 | 600 | USD |
| AWS Active Custom | 60 | 300 | 1,200 | USD |
| Google NMT | 20 | 100 | 400 | USD |
| Google custom model | 80 | 400 | 1,600 | USD |
| Google translation LLM (1:1 input/output) | 20 | 100 | 400 | USD |
| DeepL Pro Team (full allowance) | 24.99 | 124.95 | 499.80 | EUR |
Editorial model: characters × per-million price; excludes tax, review, terminology, layout, integration, and rework. AWS’s 2M characters/month free for 12 months and Google’s 500k free characters are not deducted; DeepL assumes each user fully uses the 1M-character allowance. Actual seats and usage are rounded. Google LLM assumes 1:1 input/output; longer output changes the price.
Why do multiple languages multiply the cost?
Google explicitly bills source characters × target languages; 1M characters into three languages is billed as 3M, not 1M.
This is the most common multilingual budget error. Translating a Chinese site into English, German, and Spanish is not “translate once and copy”; it is three target-language translations. Google’s batch document page states characters × target languages; AWS bills translated characters, and each target language is normally a separate request. Confirm in the contract/invoice whether each target language is additive.
| Option | 1M source characters × 3 languages | Billed as 3M characters | Note |
|---|---|---|---|
| AWS standard text | 3M | USD 45 | Confirm each target language separately |
| Google NMT | 3M | USD 60 | Official source × target rule |
| Google translation LLM (1:1) | 3M | USD 60 | Input and output metered separately |
| DeepL Pro Team | 3M | EUR 74.97 | Modelled as 3 users at full 1M allowance |
Editorial model; excludes tax, review, terminology, layout, and rework. For one target language, use 1M characters; do not multiply a single-language price by page count.
Why is the API fee not the translation budget?
Terminology, review, layout, placeholders, certifications/units/legal wording, and rework are the main lines; the API only produces a draft.
| Cost line | Included in API price? | How to meter | Common omission |
|---|---|---|---|
| Translation API/subscription | Yes | Characters × price; users × monthly fee | Target-language count, input/output split |
| Terminology/style guide | No | Term entries, maintenance hours | Product names, models, units, certification names diverge |
| Human review | No | Characters ÷ review speed × labour rate | Assuming MT can be published directly |
| Layout/DTP | No | Pages/files, languages | German/Spanish expansion breaks layout |
| Placeholders/tags | No | Variable, link, HTML tag checks | Links/variables break after translation |
| Integration/write-back | No | CMS/API, queues, retries | Translating text but not updating the site |
| Maintenance/versioning | No | Updates, retranslation, regression | Only the Chinese source is updated |
Editorial guidance, not an accounting taxonomy. Follow corporate finance policy, vendor contracts, and actual invoices.
How much human review does 1M characters need?
If 30% needs full review, at 2,000 characters/hour and RMB 150/hour, review is about 150 hours and RMB 22,500 — far above the USD 15–20 standard NMT API fee.
| Review-share assumption | Characters to review | Review hours (2,000 chars/hour) | Labour cost (RMB 150/hour) | Fit |
|---|---|---|---|---|
| 10% | 100,000 | 50 hours | RMB 7,500 | Internal reference, low risk |
| 30% | 300,000 | 150 hours | RMB 22,500 | Public product pages, general marketing |
| 100% | 1,000,000 | 500 hours | RMB 75,000 | Certifications, legal, safety, quotes |
1 million source characters, one language. Review share, 2,000 characters/hour, and RMB 150/hour are editorial assumptions, not an industry standard or quote. API fees are in USD and labour in RMB; they are listed separately and not converted or added. Replace them with your own review speed, labour rate, and risk tiers.
How to choose: API, subscription, or custom model?
Use standard API for high-volume internal content; API + terminology + human review for public pages and product claims; evaluate custom models or a subscription workspace for fixed terminology/brand voice.
| Scenario | Start with | Why | Do not ignore |
|---|---|---|---|
| Internal email/tickets/knowledge drafts | Standard NMT API | Low unit price, batchable, not every line needs full review | Sensitive data, region, terminology |
| Public website/product pages/white papers | NMT + glossary + human review | Machine draft plus human gatekeeping | Units, certifications, legal wording, links |
| Fixed terminology/brand voice | Custom model/glossary/translation memory | Reduces repeated edits and style drift | Training data, training cost, maintenance |
| Teams translating inside a UI | DeepL Pro subscription | Included characters, glossary, team admin | Compare per user/full allowance; not an API |
| Document layout/large file batches | Document translation (per page/character) | Preserves formatting, less DTP | PDF/PPT layout, text in images |
| Strict compliance/data residency | Regional cloud/private/contract | Data boundary, audit, permissions | Do not compare price alone; require DPA/security review |
Editorial guidance, not a vendor ranking. Start with 1–2 target languages and 1–2 page types; measure quality and review hours on the same glossary and rubric before expanding.
What should procurement and acceptance measure?
Measure terminology consistency, numbers/units/certifications, omissions/additions, placeholders, review hours, and cost per 1,000 characters — not BLEU or a vendor demo alone.
| Acceptance item | Suggested definition | Evidence |
|---|---|---|
| Terminology consistency | Product/model/material/certification match to glossary | Glossary, sampling record |
| Numbers/units/dates | Values, units, dates, currency checked line by line | Comparison sheet, two-person check |
| Omissions/additions | Paragraph/sentence omissions, additions, misalignment | Bilingual comparison, labelled sample |
| Placeholders/tags | Variables, links, HTML tags intact | Automated check, pre-publish review |
| Review hours | Review minutes/hours per 1,000 characters | Time record, sampling |
| Cost per 1,000 characters | (API/subscription + review + layout) ÷ 1,000 characters | Invoice, hours, layout record |
| Data boundary | Region, retention, training use, permissions, DPA | Contract, configuration, audit logs |
| Version/write-back | Target languages updated when source changes | CMS record, regression report |
There is no universal “acceptable” score; certifications, legal, safety, and quotes are high-risk and need human confirmation. Machine-translation scores do not replace business acceptance.
Who should read this — and who should not?
For teams building multilingual websites, brochures, manuals, or support content. Not for a few dozen sentences, no glossary, or a project that refuses human review.
Read if: you work in marketing, export, localization, product, IT, or procurement; you are comparing AWS Translate, Google Cloud Translation, DeepL Pro, or Azure Translator; you need to compare per-character API, per-user subscription, and per-page document pricing, and estimate review. Skip if: you have only a few dozen sentences and expect free tiers to solve production; you have no glossary/style guide; you expect MT to publish certifications/legal/safety content directly; you treat the API price as the whole translation budget. Supports: unit conversion, multilingual multiplier, 1M/5M/20M-character models, review TCO, scenario selection, and acceptance metrics.
Lingyi does not resell translation APIs or claim any engine is “most accurate”. For multilingual sites, glossaries, CMS write-back, and launch acceptance, see multilingual manufacturing websites and website cost breakdown; to scope a translation budget, glossary, and acceptance checklist, talk to us.
See our separate notes on AI coding-assistant TCO and document OCR cost; this article only covers translation API/subscription character costs and review.
FAQ
How do I compare an API with DeepL Pro?
APIs bill per character on demand, suitable for system integration and batch content; DeepL Pro bills per user/month with included characters, suitable for teams working in the UI. Divide the subscription price by actual characters used to compare with a per-million API price; at low utilisation, the subscription’s effective per-character cost is higher.
Why do multiple languages multiply the cost?
Each target language is a separate translation. Google explicitly bills source characters × target languages; 1M characters into three languages is billed as 3M. AWS bills translated characters, and each target language is normally a separate request; confirm additivity in the contract/invoice.
Is Google LLM translation cheaper than NMT?
Google translation LLM is USD 10/million input + USD 10/million output, about USD 20/million at 1:1 input/output, close to NMT’s USD 20/million; longer output costs more. Do not look at input price alone, and do not assume LLM is more accurate — test on the same glossary and sample.
How much human review does 1M characters need?
It depends on risk. If 30% needs full review at 2,000 characters/hour and RMB 150/hour, that is about 150 hours and RMB 22,500; 100% review is about 500 hours and RMB 75,000. The API may be only USD 15–20; review is the main cost. These are editorial assumptions, not a quote.
What does 1M characters cost per month?
Standard NMT: AWS about USD 15, Google about USD 20; Google custom about USD 80; DeepL Pro Team at full allowance about EUR 24.99 (one user/1M characters). Add terminology, review, layout, and integration; the table is an editorial model, not a quote.
How should we accept translation quality?
Use the same glossary, rubric, and sample, and check terminology, numbers/units/certifications, omissions/additions, placeholders/tags, review hours, and cost per 1,000 characters by target language. Do not rely on BLEU or a vendor demo; certifications, legal, safety, and quotes need human confirmation.
Method and limitations
This article uses only official pricing/product pages accessible on 31 October 2026. It does not use private contracts, customer invoices, or vendor internal data. Prices, free tiers, models, character counting, and subscription terms can change. The article has no actual enterprise agreement, tax, FX, review, or layout data, so every monthly cost is an editorial model, not a quote. A public DeepL API-specific per-character price was not obtained; only the Pro subscription allowance is used as a comparison. Azure Translator S1 values rendered dynamically and are excluded from the numeric comparison. EUR and USD are not converted; target-language count, review share, and labour rate are assumptions.
Sources and basis
Data cut-off: 31 October 2026. The following are official pricing/product pages; prices, allowances, models, and terms follow the original page and contract.
- AWS, Amazon Translate pricing: aws.amazon.com/translate/pricing. Standard/batch/real-time document/Active Custom per-million-character prices and free tiers.
- Google Cloud, Cloud Translation pricing: cloud.google.com/translate/pricing. NMT, custom model, LLM, and document prices; source × target language and input/output metering.
- DeepL, Pro pricing: deepl.com/en/pro-api. Individual/Team/Business subscription prices, included characters, glossary/SSO/translation memory; API-specific price not obtained.
- Microsoft Azure, Translator pricing: azure.microsoft.com/pricing/details/cognitive-services/translator. F0 2M free characters/month; S1 values rendered dynamically and not captured.