Translate text between 67 languages in your browser: eight English directions use fast per-pair Opus MT, the rest use the multilingual NLLB-200 model.
All translation happens in your browser. Your text never leaves your device. Models download once and your browser reuses them on later visits.
Runs entirely in your browser. Your data never leaves your device.
When you click Translate, three things happen — all locally, all in your browser:
| Direction | Model |
|---|---|
| English → Español | opus-mt-en-es (Opus MT) |
| Español → English | opus-mt-es-en (Opus MT) |
| English → Deutsch | opus-mt-en-de (Opus MT) |
| Deutsch → English | opus-mt-de-en (Opus MT) |
| English → Nederlands | opus-mt-en-nl (Opus MT) |
| Nederlands → English | opus-mt-nl-en (Opus MT) |
| English → Indonesia | opus-mt-en-id (Opus MT) |
| Indonesia → English | opus-mt-id-en (Opus MT) |
| Português ↔ English | nllb-200-distilled-600M (NLLB-200) |
| All other pairs across the 67 listed languages | nllb-200-distilled-600M (NLLB-200) |
No. This AI translator runs entirely in your browser using WebAssembly. Your text never leaves your device. Models are downloaded once from CDN and cached locally.
We support 67 languages in the picker. Eight high-frequency directions (English–Spanish, English–German, English–Dutch and English–Indonesian, both ways) use a dedicated per-pair Opus MT model (~30MB each). Every other combination is served by one large multilingual model, Meta NLLB-200-distilled-600M (~600MB). Note that accuracy and download size differ between the Opus MT and NLLB-200 paths.
Per the upstream model cards, Opus MT models are fine-tuned on millions of sentence pairs from OPUS datasets and produce fluent translations for general-purpose text (news, emails, documentation). The NLLB-200 path uses one shared multilingual model, so quality varies by language pair; we publish no measured scores for this tool.
No. The model for a language direction is downloaded once (the Opus MT pair models are quantized at ~30MB); your browser then reuses the files from its standard HTTP cache (the Cache API), so later visits skip the big re-download. This is not a full offline mode — there is no service worker, so the page itself still needs a connection.
The model accepts up to 512 tokens per translation (approximately 400 words). For longer texts, use batch mode: separate paragraphs with blank lines.
If you just finished with AI Translator, the natural next steps are AI Subtitle Generator, AI Text Summarizer, AI Paraphraser, or browse every tool in Local AI Tools.