DeepL

DeepL is purpose-built neural machine translation rather than a general-purpose LLM. It does one job, does it fast, and does it across 240+ languages and variants — the widest language coverage in the catalog. It is also considerably more steerable than “machine translation” suggests, which is the part most comparisons miss.

​

Models

​

Steering controls

The combination of glossaries, translation memory, and custom instructions is what makes DeepL viable for professional document work rather than only for bulk gisting.

​

Document translation

Formatting is preserved across docx, pptx, xlsx, pdf, html, txt, xliff, srt, idml, xml, json, dita, and mif. Output can optionally be converted to a different format, and a “Translated by DeepL” watermark can be applied to docx and pdf. DeepL Extended adds translation of images embedded inside documents — the difference between a translated PDF whose diagrams, charts, and screenshots are still in the source language and one that is genuinely localized. It is marked experimental, so review output before it reaches a deliverable.

​

Where it fits in a workflow

DeepL is the right answer for high-volume professional document translation where the content is functional rather than expressive: manuals, policies, specifications, internal documentation, and knowledge bases. A common configuration overrides DeepL at Folder level for bulk documentation while leaving Agentic Localization in place as the Global Workflow default for customer-facing content.

​

Trade-offs

• No AI Dubbing support.
• NMT, not an LLM. Custom instructions are capped at 300 characters each and it will not reason about ambiguous source text or adapt cultural references the way the frontier models do. For marketing, narrative, or brand-voice content, that gap is visible.
• Image translation inside documents is a beta capability upstream.

​

Reference

• Translation provider catalog
• DeepL supported languages
• DeepL document translation