Bulk Text Translator (50 Languages, No API Key, Private)

Translate thousands of product descriptions, support replies, reviews, UI strings or documents into 50 languages, and into several target languages in one run, at a fraction of the cost of commercial translation APIs.

It runs open-source neural machine translation (Argos Translate packages of OPUS-MT models, via CTranslate2) inside the Actor. You don't need a Google or DeepL key, and your texts are never sent to a third-party service.

Highlights

• 50 languages: Arabic, Chinese (simplified and traditional), Dutch, French, German, Hindi, Indonesian, Italian, Japanese, Korean, Polish, Portuguese (PT and BR), Russian, Spanish, Turkish, Ukrainian, Vietnamese and more
• Any-to-any: pairs without a direct model are routed through English (shown as pivot: "en")
• Keeps line breaks and paragraphs; long texts (up to 20k characters) are split by sentence
• One text into many languages per run (targetLanguages: ["es","de","fr","ja"])
• ~$0.60 per million characters, compared with ~$20/M for typical cloud translation APIs

Input

Field	Default	Notes
texts	—	Up to 5,000 texts, 20,000 characters each
sourceLanguage	en
targetLanguages	["es"]	One or more
beamSize	2	1–4 (higher is a bit better and slower)

Output (one row per text × target language)

{
  "index": 0, "sourceLanguage": "en", "targetLanguage": "es",
  "sourceText": "Hello! Your order has shipped and should arrive on Tuesday.",
  "translatedText": "¡Hola! Su pedido ha enviado y debe llegar el martes.",
  "characters": 59, "billedUnits": 1, "pivot": null, "error": null
}

Pricing (pay per event)

• $0.0006 per started 1,000 source characters, per target language (translated-1k-chars). That's $0.60 per million characters.
• A 300-character text into 3 languages = 3 units = $0.0018.
• Empty or too-long texts and unavailable pairs are returned free. The Actor stops before going over your run's maximum cost.

Quality: be realistic

This is good machine translation, suitable for gist, internal use, catalogs, user-generated content and first drafts. For legal, medical or marketing copy, have a human review it. Quality is best for pairs with English and lower for pivoted pairs.

Credits and licenses

Models: OPUS-MT by Jörg Tiedemann and Santhosh Thottingal, University of Helsinki (CC-BY 4.0), packaged by the Argos Translate project (MIT). Runtime: CTranslate2 (MIT), SentencePiece (Apache-2.0), sacremoses (MIT), subword-nmt (MIT).

---

Actively maintained. Issues answered within 24 hours on the Issues tab.