DeepL Pro & API vs Google Cloud Translation

DeepL Pro & API

Own engine, 126 languages with a published per-language feature matrix, glossaries that adapt terminology grammatically and can restructure whole sentences, formality plus style rules and custom instructions, and document translation that keeps layout across Word, PowerPoint, PDF and InDesign, scanned PDFs included. Account-scoped adaptive translation, customer training data confined to the customer, and a glossary generator built from the customer's own sentence pairs point upward; what holds it down is that we found no public information on quality estimation or scoring, and the blind-test quality claim carries no published method.

Google Cloud Translation

189 languages with a published list, glossaries applied at request time from TSV, CSV or TMX files, document translation that preserves layout for Office and PDF, AutoML custom models and an LLM-based adaptive mode that learns from your own datasets, plus quality prediction scores in the Translation Hub tiers. We found no public information on a published quality evaluation with its method, nor a stated rule that routes weak segments to human review, so it stays below the top band.

DeepL Pro & API

Translation memories upload as TMX, are shared with the team and switched on in the translator; glossaries are central, shared, merged and priority-ordered, with analytics and role-based permissions alongside. But this is an engine with memory, not a review pipeline: we found no public information on review or approval stages, automated QA checks, in-context editing, per-language roles or ordering translators — the terms instead let customers connect external CAT tools to the paid API.

Google Cloud Translation

The Translation Hub tiers are named with translation memory, human review and machine translation quality prediction, and glossaries get full create-read-update-delete through the API. We found no public information on QA checks, review stages, roles per language, in-context editing or inviting translators, so what is documented reads as a review flag on a machine translation pipeline rather than a managed process.

DeepL Pro & API

A documented REST API with its own authentication scheme, a deprecation notice and granular status codes, plus named connectors for Microsoft 365, Google Workspace, Zendesk, Adobe Experience Manager, Google Drive and SharePoint, CAT-tool integration, MCP connections to ChatGPT, Claude and Cursor, and browser extensions for Chrome, Edge and Firefox. The gaps a developer will hit early: we found no public information on Git repository sync, CI pipelines, SDKs, a command-line tool, webhooks or over-the-air delivery, and the format list leans office- and media-heavy (Word, PowerPoint, Excel, PDF, InDesign, SRT) with Contentful JSON as the only software-format nod.

Google Cloud Translation

A documented REST API with client libraries, published quotas and limits, batch jobs reading and writing through Cloud Storage, and document formats from DOC and DOCX to PDF, PPT, PPTX, XLS and XLSX. We found no public information on Git or CI sync, CMS or design connectors, webhooks, or software localization formats such as XLIFF or JSON.

DeepL Pro & API

For the paid tiers the pages are unusually specific: texts deleted after processing, never used to train DeepL's models, a 72-hour cap on encrypted debug storage, 90-day deletion of any requested storage after the agreement ends, TLS and BYOK encryption, SSO over SAML and OIDC, audit logs, role-based permissions, and ISO 27001, SOC 2 Type II and BSI C5 Type 2 with a downloadable audit verdict and a published data processing agreement. Two caveats I would read twice before signing: the free API tier reserves the right to store content perpetually and the free translator trains its networks on user texts, and the voice path names ElevenLabs as a subprocessor.

Google Cloud Translation

The data-usage statements are unusually concrete: text is 'held briefly in-memory' and not stored on servers, is not used to train Google's translation features, is not shared with third parties, and Google claims no ownership of it; glossaries are encrypted in your own Cloud project behind fine-grained IAM, under a published Cloud Data Processing Addendum. We found no public information on ISO 27001 or SOC 2 certification with its scope, audit logs, or a dedicated deployment option, which keeps this below the top band.

DeepL Pro & API

The translation memory exports as TMX through an asynchronous API job with a short-lived download URL, imports from TMX files up to 1 GB, and content stored at the customer's request is deleted 90 days after the agreement ends — that much is documented. The rest is not: we found no public information on exporting glossaries or termbases as TBX or CSV, exporting projects with comments and history, or any self-hosted edition, so the memory leaves as TMX and how the glossaries leave is an open question.

Google Cloud Translation

Glossaries import as TSV, CSV or TMX 1.4, their entries can be listed back through the API, and the glossary input file lives in your own Cloud Storage bucket. We found no public information on translation memory export, a complete project export, or the return and deletion of data when the contract ends — the assets and terms that decide what leaving costs.

DeepL Pro & API

DeepL SE is registered in Cologne, translation runs exclusively on DeepL's own infrastructure in the EEA, a data processing agreement is published, and ISO 27001, SOC 2 Type II and BSI C5 Type 2 are stated. But the captured pages give different pictures of processing location — the privacy policy says EEA while the license terms reserve DeepL's right to process on third-party cloud infrastructure at locations it determines — and the chain names Cloudflare's San Francisco CDN, ElevenLabs backup voices and Apryse's PDF SDK, the latter two without a stated location.

Google Cloud Translation

A published data processing addendum and an Irish contracting and service entity are evidenced, and the Advanced API offers regionalized endpoints configurable for data location. But the Basic and AutoML endpoints are global, where the terms say 'data cannot be configured to remain in a specific region for processing', the parent is Google LLC of California, and we found no subprocessor locations among the captured facts — so sovereignty is partial and depends on choosing the right endpoint.

DeepL Pro & API

Seat prices for Individual, Team and Business (€7.49, €24.99 and €49.99 per user/month, billed annually) are published with character and file-translation limits, and the API side is fully metered: a one-time free credit of 1 million characters, Growth at €23.80 per month billed annually plus "€22.00 / 1,000,000 extra characters", the Write Pro add-on at €6.99 per user/month, and VAT treatment and trial auto-conversion stated. What a buyer cannot compute: Enterprise is custom pricing through sales, "unlimited characters" rests on an undefined fair usage policy, and we found no public information on minimum terms or the monthly-billing price of the Pro tiers.

Google Cloud Translation

Per-character prices by model are public — $20 per million for NMT, $10 for the LLM, $25 for adaptive, $80 for custom with volume bands down to $30 — alongside per-page document and Translation Hub prices, $45-per-hour custom training capped at $300 per job, 500,000 free characters a month, $300 in new-customer credits, and one- and three-year committed discounts. A buyer can compute an annual bill from public pages; we found no public information on VAT treatment or support fees, so it stops just short of full computability.