← all SaaS
Self-host DeepL
Machine translation API ·
Category: AI / LLM tooling
DeepL is the German translation service known for translation quality that beats Google Translate on European-language pairs. The self-hostable replacements all run open neural-MT models on your own hardware; quality on common pairs (EN ↔ DE/FR/ES) is now within striking distance of DeepL, with the gap widening on lower-resource pairs.
DeepL pricing anchor: Free 500k chars/mo; Starter $8.74/user/mo; Advanced $28.74/user/mo; Pro $57.49/user/mo.
- GitHub
- ★ 14.3k · last commit 4d ago · 118 open issues
- License
-
AGPL-3.0 - Setup time
- 10min docker run (model downloads on first start)
- Monthly cost
- $10-30/mo VPS — CPU-only works for low traffic; GPU recommended above 100 req/min.
Migration sketch. `docker run -ti --rm -p 5000:5000 libretranslate/libretranslate --load-only en,de,fr,es`. Models (Argos OpenNMT) download on first start; subsequent runs use the cached models. The HTTP API is DeepL-compatible-ish — most clients only need to swap the base URL: `POST /translate {q, source, target, format}`. For DeepL parity at scale, deploy the Argos packages on a GPU box with batch translation.
Good fit forWeb apps that want a self-hosted translation API for user content (UGC translation, support ticket I18n).
Weak atQuality on rare pairs (EN ↔ Slavic, EN ↔ Asian) is below DeepL; for high-stakes translation this still matters.
- GitHub
- ★ 5.96k · last commit 2w ago · 150 open issues
- License
-
MIT - Setup time
- 5min `pip install argostranslate` + download models
- Monthly cost
- $0 on a workstation; ~$10/mo for a small VPS for batch jobs.
Migration sketch. `pip install argostranslate`. In Python: `import argostranslate.package, argostranslate.translate; argostranslate.package.update_package_index(); pkg = next(p for p in argostranslate.package.get_available_packages() if p.from_code=='en' and p.to_code=='de'); argostranslate.package.install_from_path(pkg.download()); print(argostranslate.translate.translate('Hello world', 'en', 'de'))`. Argos is the engine LibreTranslate wraps — pull it in directly for offline / scripted translation pipelines.
Good fit forBatch translation pipelines (subtitle files, document translation jobs) where you don't need an HTTP service.
Weak atNo HTTP API by default; you build the service layer if you need one (or use LibreTranslate).
- GitHub
- ★ 812 · last commit 3mo ago · 57 open issues
- License
-
MIT - Setup time
- 30min — pull Marian + a model from Hugging Face
- Monthly cost
- $15-50/mo VPS — Marian is C++ and fast, but GPU helps for throughput.
Migration sketch. Use the Hugging Face transformers wrapper for the easy path: `pip install transformers torch && python -c 'from transformers import pipeline; t=pipeline("translation", model="Helsinki-NLP/opus-mt-en-de"); print(t("Hello world"))'`. For production, deploy the underlying Marian-NMT C++ binary with a thin HTTP wrapper. Helsinki-NLP publishes models for ~1000 language pairs — easily the broadest open coverage.
Good fit forResearch, low-resource language pairs, and teams that want to fine-tune translation models for a specific domain.
Weak atBring your own service layer; quality varies by pair (well-resourced pairs are excellent, others vary).
In a terminal? npx os-alt deepl prints this table —
how the CLI works →