1 credit = 1 translated word. We count whitespace-separated tokens in the source text. Before each translation, we show you the exact cost and ask you to confirm. Editing and exporting are always free.
Pack credits never expire — buy a pack once, use it whenever you want. Subscription credits renew every month (unused credits roll over up to twice your monthly allowance).
If you translate regularly, a monthly subscription gives you more credits for the same price and renews automatically — cancel anytime, no commitment. If you translate occasionally, one-time packs are simpler: pay once, and the credits never expire. You can also mix both — subscription credits are used first, then your pack credits.
Yes — every new account gets 500 credits free. Enough to translate a page or two and see if the quality matches your standards.
Unused credits are refundable within 14 days of purchase. Used credits aren't, but if our service was broken or unusable, contact us and we'll make it right.
For mainstream language pairs (English ↔ French, Spanish, German, Italian, Portuguese, Mandarin, Japanese, Arabic, etc.) the quality is comparable to a careful human first draft. For rarer pairs, expect a serviceable base you'll want to polish. You can always edit any word inline.
108 at the moment, covering every major language family — Romance, Germanic, Slavic, Indic, Semitic, Sino-Tibetan, CJK, African, Turkic, and more. Right-to-left scripts (Arabic, Urdu, Hebrew, Farsi) are rendered natively.
Yes. Before translating, pick a style preset (technical, administrative/legal, medical, academic, etc.). The AI adjusts its register and terminology accordingly.
The "Literary — classical / archaic" style preset asks the AI to use the older, formal literary register of the target language (Shakespearean English, classical Arabic fusha, classical Persian, etc.). Useful for poetry, scripture, period pieces.
PDF, DOCX, DOC, TXT, RTF, and images (JPG, PNG, WebP, HEIC). Up to 500 pages per file, 20 MB max.
Yes. If the PDF doesn't have a text layer, we automatically run OCR on each page.
PDF, Word (.docx), and Bilingual XLSX (source and target side by side). DOCX and XLSX handle all 100+ languages including Arabic, Urdu and CJK. PDF export currently best handles Latin-script languages; for non-Latin we recommend DOCX.
We preserve paragraph structure, line breaks and page boundaries — meaning the translation is readable in the same logical order as the source. We don't currently reproduce the exact visual layout (fonts, images, exact positioning) of the original PDF. We're working on this.
Yes. Documents are stored on our servers only while you work on them, encrypted at rest, and you can delete any project at any time. We never use customer documents to train AI models.
Only you. Each user account is isolated via row-level security in our database.
EU (France) for the application servers. Translation requests go to Google's regional inference endpoints.
Email + password, Google or GitHub. All three work the same.
Yes. Email us at hello@metaphras.ai and we'll delete your account and all associated data within 7 days.