10 pages of a German contract, deadline in 3 hours, and you haven’t even started. Sound familiar? A couple of years ago your only option was to sit down and translate manually. Now AI translators can produce a first draft in minutes. The only question is - which one to pick so you don’t end up redoing half the text.
I’ve tested the five most popular AI tools for document translation in 2026. No marketing fluff - just real capabilities, prices, and limitations.
What we’re comparing and how¶
For document translation, four things matter: translation quality (especially for Ukrainian), formatting preservation, file format support, and price.
Here’s the overview:
| Tool | Ukrainian Quality | File Formats | Formatting | Price (API per 1M chars) |
|---|---|---|---|---|
| DeepL | Good | DOCX, PDF, PPTX, XLSX, HTML | Excellent | $25 |
| Google Translate | Adequate | DOCX, PDF, PPTX, XLSX | Moderate | $20 |
| ChatGPT (GPT-4o/5) | Moderate | PDF, DOCX upload | Poor | ~$2.50-10 per 1M tokens |
| Claude (Anthropic) | High | PDF, DOCX upload | None | ~$3-25 per 1M tokens |
| Microsoft Translator | Adequate | DOCX, PDF, PPTX, XLSX | Good | $10 |
Let’s dig into each one.
1. DeepL - best for formatted documents¶
DeepL remains the gold standard for document translation in 2026. The reason is simple - it doesn’t just translate text, it gives you back a file with the original formatting intact. Tables, headers, bold text, numbered lists - everything stays in place.
What it does: - Translates DOCX, PDF, PPTX, XLSX, HTML, TXT, and even subtitles (SRT) - Best formatting preservation among all competitors - Glossary support in Pro version - you can define how specific terms should be translated - 33 languages, including Ukrainian (since September 2022)
Pricing: - Free: 1,500 characters per request, 3 files/month (up to 5 KB each) - Starter: $10.49/month - unlimited text, 5 files - Advanced: $34.49/month - 20 files, glossary - Ultimate: $68.99/month - 100 files, priority support - API Pro: from $5.49/month + $25 per 1M characters
Issues: the free tier is nearly useless for real work - 1,500 characters is one paragraph. After the model update in 2025, some forum users report quality regression: “translations started sounding like AI” - ironic for a service that used to stand out for its natural output. For Ukrainian, quality is good but not at the level of DeepL’s core languages (DE, FR, ES).
More on DeepL’s Ukrainian quality in our DeepL vs Google Translate comparison.
2. Google Translate - free with the widest coverage¶
Everyone knows Google Translate, and that’s both its strength and its problem. Strength - 249 languages, free access without hard limits. Problem - many people assume Google Translate quality is “good enough,” though for professional translation that’s often not the case.
What it does: - 249 languages - more than any competitor - Free document translation via web interface with no quantity limits - Supports DOC, DOCX, PDF, PPTX, XLSX - Gemini integration improved quality noticeably in 2025-2026
Pricing: - Web interface: free, 5,000 characters per request - Cloud API: $20 per 1M characters (first 500K/month free) - Document Translation API: $0.25 per page
Issues: formatting preservation is mediocre - complex tables and headers often break. Users on Google’s support forums report specific bugs with Ukrainian - the service doesn’t always correctly recognize letters like “є”, “ї”, “ґ” and sometimes confuses Ukrainian with Russian. For legal documents, the quality isn’t sufficient.
3. ChatGPT (OpenAI) - best for creative content¶
ChatGPT (running on GPT-4o and GPT-5) isn’t a dedicated translator - it’s a general-purpose AI that happens to translate well. Its superpower is context understanding and the ability to adapt translations to match a specific style.
Marketing copy where tone and brand voice matter? ChatGPT handles it better than DeepL. A legal contract with precise terminology? That’s where it gets risky.
What it does: - Translates uploaded PDF and DOCX files (up to 512 MB) - Understands context and can adapt translation style based on instructions - You can set up a glossary right in the prompt - Works with 90+ languages
Pricing: - Free: 3 files per day - Plus: $20/month - 80 files per 3 hours on GPT-4o - API: GPT-4o from $2.50 per 1M tokens (input), GPT-5 from $10 per 1M tokens (output)
Issues: doesn’t preserve document formatting - you get translated text, not a formatted file. Sometimes “takes liberties” with translations, adding or changing things that aren’t in the original. Quality for Ukrainian is weaker than for major European languages. And there’s no translation memory - every new request starts from scratch.
4. Claude (Anthropic) - highest quality for long documents¶
At the WMT24 machine translation competition, Claude 3.5 ranked first in 9 out of 11 language pairs. In Lokalise’s blind evaluation, 78% of Claude’s translations were rated “good” by professional translators - more than GPT-4, DeepL, and Google Translate.
Claude’s main advantage is its context window of up to 200K tokens (and up to 1M via API). This means it can “keep in mind” the entire document and maintain consistent terminology from the first page to the last. For long documents, that’s a critical advantage.
What it does: - Highest translation quality among all AI tools (per independent benchmarks) - Context window up to 1M tokens - ideal for long documents - Consistent terminology throughout the text - Style, tone, and glossary can be defined in the prompt
Pricing: - Free plan: limited number of messages - Pro: $20/month - API: Sonnet from $3 per 1M tokens (input), Opus from $15 per 1M tokens (output)
Issues: like ChatGPT, it doesn’t return a formatted file - just text. No specialized document translation interface. You need the API for full capability. No built-in glossary or translation memory.
Claude is what powers translation at ChatsControl - the platform uses it as its foundation, adding document formatting preservation and multiple rounds of AI quality review.
5. Microsoft Translator - the quiet corporate player¶
Microsoft Translator rarely comes up in comparisons, but it deserves attention. Especially if you’re already in the Microsoft ecosystem - Word, PowerPoint, and Outlook have built-in integration.
What it does: - 100+ languages, including Ukrainian - Good document formatting preservation - Can translate text inside images in Word documents (2025 feature) - Custom models for domain-specific terminology via Azure
Pricing: - Free tier for testing - Standard API: $10 per 1M characters - cheapest among the “big five” - Custom models: $40 per 1M characters
Issues: less polished interface than DeepL. Ukrainian quality is middle-of-the-road - better than Google for some pairs, worse than DeepL. Primarily focused on enterprise clients, not the most convenient option for freelancers.
New players in 2025-2026: specialized tools¶
Beyond the “big five,” specialized services have appeared that address specific pain points:
Smartcat - enterprise platform supporting 280+ languages and 50+ file formats. AI agents that translate, localize, and adapt content. Great for agencies and large-scale projects with repetitive content.
Taia - combines AI speed with human quality assurance. Converts PDFs to Word, translates while preserving structure, outputs an editable DOCX. Pricing from $0.54 to $2.70 per word.
Doclingo, TranslatesDocument - specialize in preserving document layout. If it’s critical that your translation looks identical to the original - these tools are worth trying.
Which tool for which task¶
| Task | Best Choice | Why |
|---|---|---|
| Translate .docx with formatting | DeepL or ChatsControl | DeepL preserves formatting, ChatsControl adds AI quality review |
| Quickly understand a foreign language document | Google Translate | Free, no limits, 249 languages |
| First draft for MTPE post-editing | Claude or DeepL | Claude gives highest quality, DeepL needs fewer edits |
| Marketing copy translation | ChatGPT | Best at adapting tone and style |
| High-volume corporate documents | Microsoft Translator | $10 per 1M characters, Office integration |
| Legal document translation | None as a final version | Human review by a translator is required |
| Documents for visa or Germany | Sworn translator | Machine translation isn’t accepted by authorities |
MTPE: how professionals actually use AI¶
87% of freelance translators already work with MTPE (Machine Translation Post-Editing) - regularly or occasionally. It’s not about “replacing translators” - it’s a tool that doubles productivity.
Here’s how it works in practice: you run a document through an AI translator, get the first draft, then edit it. With a good base (Claude or DeepL), post-editing takes 30-50% less time than translating from scratch. A translator using MTPE handles 4,000-5,000 words per day instead of 2,000.
DeepL claims their translations need half as many edits as Google and three times fewer than ChatGPT. Claude delivers the highest overall quality according to professional translator evaluations. For MTPE, both are an excellent foundation.
FAQ¶
Which AI translator is best for Ukrainian?¶
According to independent WMT24 benchmarks, Claude (Anthropic) shows the highest translation quality among AI tools. DeepL works well for DE-UK and EN-UK pairs, especially with legal and business texts. Google Translate is fine for quickly grasping the meaning of a text, but quality isn’t sufficient for professional use.
How much does AI document translation cost?¶
The range is wide: from free (Google Translate web version) to $25 per 1M characters (DeepL API Pro). Microsoft Translator is $10 per 1M characters. ChatGPT Plus costs $20/month with a fixed limit. For small volumes (a few documents per month), free tiers from DeepL or Google will do.
Do AI translators preserve document formatting?¶
DeepL preserves formatting best - tables, fonts, and layout stay intact. Microsoft Translator also does a decent job. Google Translate keeps basic formatting, but complex layouts can break. ChatGPT and Claude don’t return a formatted file at all - just the translated text. ChatsControl solves this problem by preserving full .docx formatting.
Can you use AI translation for official documents?¶
No. For official documents for Germany - visas, diploma recognition, family reunification - you need a sworn translation done by a human with an official seal. Machine translation, even perfectly edited, has no legal standing and won’t be accepted by any authority.
What’s MTPE and should translators use it?¶
MTPE (Machine Translation Post-Editing) is when AI creates the first draft and a translator edits it. 87% of freelancers already use this approach. Time savings are 30-50%, and productivity jumps from 2,000 to 4,000-5,000 words per day. The best bases for MTPE are DeepL (with glossary) and Claude (highest overall quality).