DeepL vs Google vs Azure Translation: monthly text and document costs
Translation APIs don't all bill the same thing. Some meter raw-text characters, some charge per document page, and some apply a minimum character charge per file. Two quick traps to avoid: Azure's free F0 tier and paid S1 tier are separate purchases, and DeepL's annual allowance shouldn't be treated as a monthly grant.
For a comparison with token-priced language models, use the translation API versus LLM workload guide. DeepL, Google Translation or an LLM: API Cost per Job.
Choose the job before choosing the cheapest rate
Translation pricing trips up a lot of engineering teams, and it's easy to see why. A real-time support chat sends short text snippets. An ecommerce localization pipeline translates product descriptions into six languages simultaneously. A legal team needs formatted multi-page PDF contracts translated with layouts preserved. Each of those jobs runs on a completely different billing meter. Whatever you pick, budget human review and terminology checking right alongside your API consumption.
A quick note on the numbers: all reference rates below were verified on 25 October 2026 in USD before sales tax. DeepL uses official US monthly Growth pricing; Google Cloud Translation uses standard NMT pay-as-you-go; Azure Translator uses East US S1 commercial meters. Custom models, cloud storage, and enterprise volume discounts are excluded.
| API route | Allowance and cash commitment | Text-translation meter |
|---|---|---|
| DeepL Developer | One-time 1 million characters, not a monthly subscription allowance | Testing allowance; do not forecast a renewing free production tier |
| DeepL Growth, monthly | $32.50/month; 1 million characters/month included | $27.50 per additional million; 50 million/month plan usage limit |
| Google standard NMT | $10 monthly credit equivalent to 500,000 characters | $20 per million source characters per target language |
| Azure F0 | Up to 2 million eligible characters/month on the free resource | Separate free tier; no assumed automatic paid overage |
| Azure S1 | Paid consumption; no F0 allowance deducted in this model | $10 per million standard text characters |
What to keep in mind on DeepL: the annual Growth plan provides a 12 million character yearly allowance rather than monthly increments. That can be handy for seasonal workloads, but our worked examples use the monthly Growth plan ($32.50/month, $27.50 per additional million characters up to 50 million/month) so your cash outlay lines up with what you actually use each month.
Two more gotchas worth knowing. DeepL API keys don't grant access to the web translation interface, which requires separate per-user licenses. And on Azure, the F0 free tier (2 million characters/month) lives on a separate free resource, so it doesn't automatically deduct from an active S1 commercial invoice.
Monthly plain-text budgets
Let's start with the simple case: standard plain-text translation for one target language with zero retries. Volumes represent source characters, not words or API calls. And if you translate into a second target language, you double the source characters processed.
| Monthly source characters | Azure F0 / paid S1 | Google NMT | DeepL Growth monthly |
|---|---|---|---|
| 1 million | F0: $0 if eligible; S1: $10 | (1 − 0.5) × $20 = $10 | $32.50 |
| 10 million | F0 alone cannot cover it; S1: 10 × $10 = $100 | (10 − 0.5) × $20 = $190 | $32.50 + 9 × $27.50 = $280 |
| 100 million | S1 reference: 100 × $10 = $1,000 | (100 − 0.5) × $20 = $1,990 | Above Growth’s monthly cap; Enterprise quote needed |
At 10 million characters, Azure S1 costs $100, while Google NMT comes to $190 (after the $10 credit) and DeepL Growth costs $280. Azure saves $90 over Google, which looks great on paper. The practical catch: at a standard $45/hour editorial rate, just two hours of human terminology correction wipes out that entire $90 savings. So try not to pick a translation engine on API pennies alone. Test output quality on your own domain vocabulary first.
If you translate into two languages, a 10M source character batch becomes 20M billed characters. Google's $10 monthly credit applies once across your whole project, not per target language. And if you go with Google Cloud Translation LLM models, input and output tokens are billed separately under generative model rates.
Formatted documents change the meter
Document translation plays by a different set of rules. DeepL enforces a 50,000-character minimum per formatted file (Word, PowerPoint, Excel, or PDF), even if the document contains only 500 words. Plain text, HTML, SRT, and XLIFF files are billed on actual character counts without that minimum. The upshot: uploading 100 short PDFs on DeepL burns through quota fast.
Google Cloud charges a flat $0.08 per page for DOCX, PPT, and PDF files, while Excel documents bill under text character rates. Azure S1 bills standard documents at $15 per million document characters, with separate pricing for scanned image documents.
To see how these models compare in practice, picture a workflow processing four-page PDFs averaging 12,000 characters each into one target language:
| Monthly PDF files | Azure S1 document characters | Google NMT document pages | DeepL monthly Growth |
|---|---|---|---|
| 100 | 1.2M × $15 = $18 | 400 × $0.08 = $32 | 5M billed characters: $32.50 + 4 × $27.50 = $142.50 |
| 1,000 | 12M × $15 = $180 | 4,000 × $0.08 = $320 | 50M billed characters: $32.50 + 49 × $27.50 = $1,380 |
| 10,000 | 120M × $15 = $1,800 | 40,000 × $0.08 = $3,200 | 500M billed characters exceeds Growth’s monthly limit |
Here's where the file minimum bites: 100 PDFs on DeepL consume 100 × 50,000 = 5,000,000 billable characters, totaling $142.50. At 1,000 PDFs, DeepL hits 50M characters (its monthly plan ceiling) and bills $1,380. Azure S1 handles the same 1,000 PDFs for $180, while Google charges $320 (4,000 pages × $0.08).
One workaround people reach for: extract the text from the PDFs and send raw text to DeepL, which costs only $38 for 1.2M characters. It's a real saving on the API bill, but rebuilding complex tables and multi-column formatting by hand can cost far more in design labor. Make sure formatting preservation is part of your total cost math.
What to measure before committing
My advice is to run a small pilot with your actual documents. Track characters billed, formatting fidelity, glossary support, and turnaround time. For high-volume enterprise translation, get custom committed-use quotes from DeepL, Google, or Microsoft rather than stretching self-service tiers beyond their published limits.
If you're building a hybrid human-in-the-loop workflow, our translation and post-editing guide is a good next stop. And the interactive worksheet below models DeepL Growth document billing with exact per-file minimums, so you can plug in your own numbers.
Budget formatted files on DeepL monthly Growth
Defaults: 100 PDF files, one target language, a 50,000-character minimum per file, 1M monthly allowance and $32.50 base fee. Use at least 50,000 per Word/PowerPoint/Excel/PDF file. The worksheet does not enforce the per-file minimum or the 50M/month plan cap; compare demand with both before using the result. Excludes human review and tax.
Add human review to the translation budget →