DeepL Pro & API vs Google Cloud Translation

DeepL Pro & API

This is close to the whole answer for my contracts and manuals: an own engine with a published list of 126 languages and per-language features, glossaries that enforce house terminology with grammatical adaptation and do-not-translate terms, a formal/informal setting, and file translation that keeps the layout of Word, PowerPoint and PDF — scanned PDFs included — plus InDesign, images and Excel. What holds it back: I found no public information on automatic quality estimation that would route weak segments to review, and the blind-test quality claim comes without a published method.

Google Cloud Translation

For my desk this covers the essentials: 189 languages with any-to-any pairs, glossaries enforced during translation (TSV, CSV or TMX import, with case handling and fine-grained permissions), and document translation that preserves the original formatting and layout for Word, PowerPoint and PDF, including scanned PDFs — though the documentation warns that content inside text boxes is not translated. Custom AutoML models and LLM-based Adaptive Translation that works from our own datasets are available, but custom models cover only 53 language pairs via English, and we found no published quality evaluation with its method and no formality or tone setting.

DeepL Pro & API

As an engine rather than a managed process it does what I want without forcing a TMS on me: a translation memory uploaded as TMX, shared with the team and reused for similar texts, a glossary manager with shared glossaries, merging and priority order, plus team administration and a claimed workflow feature. I found no public information on review or approval stages, automated checks on placeholders or terminology, or inviting translators and agencies, so it stays low on a scale built around review workflow.

Google Cloud Translation

The glossary side is genuinely manageable — glossaries and their entries can be created, updated, listed and deleted through the API, though without version control — and the managed Translation Hub adds translation memory, human review and machine translation quality prediction scores on its advanced tier. Beyond those one-line feature mentions we found no description of configurable review stages, automated checks for placeholders or terminology, comments, or translator roles per language. As someone who wants documents back rather than a project platform that suits me, but the review machinery is thin on the documented pages.

DeepL Pro & API

Files in and out is my world: PDF, Word, PowerPoint, Excel, HTML, TXT, SRT, JPG, PNG and InDesign with formatting preserved, Microsoft 365 and Google Workspace add-ins, Chrome, Edge and Firefox extensions, Google Drive, SharePoint, Zendesk and Adobe Experience Manager connectors, and a documented REST API — including translation-memory import and export endpoints with their limits spelled out. I found no public information on a CLI, SDKs, Git synchronisation, CI pipelines or webhooks, which a software team would miss more than I would.

Google Cloud Translation

A fully documented REST API with client libraries for both editions, published quotas and limits, batch jobs of up to 100 files, and input and output through Cloud Storage; the document formats are listed precisely, from DOC and DOCX through PDF, PPT and PPTX to XLS and XLSX. We found no documented sync with Git repositories or CI pipelines, no connectors or plugins for CMS, shop or design tools, and no software formats such as XLIFF, PO or JSON for the content itself.

DeepL Pro & API

The written promise I need before a draft contract leaves the building is on their pages: for Pro subscribers texts are stored only as technically needed and deleted afterward, never used to train their models, with a stated 72-hour debugging exception and deletion 90 days after contract end where storage was requested. Add TLS encryption, bring your own key, SSO with OIDC and SAML, MFA, audit logs, role-based permissions, ISO 27001 and SOC 2 Type II with its scope named, a BSI C5 attestation and a downloadable data processing agreement; I found no public information on a dedicated deployment, and the free developer API reserves perpetual storage, which I would simply not use for contracts.

Google Cloud Translation

This is the written promise I need before sending a draft contract out: Google states it does not use any sent content except to provide the service and does not use it to train or improve its translation features, that text is held briefly in-memory only and is not stored on servers, and that content is neither shared with third parties nor made public. Glossaries are encrypted with fine-grained access permissions, and a Cloud Data Processing Addendum with GDPR commitments is incorporated into the agreement. We found no statement on audit logs or ISO 27001/SOC 2 certification for this product, which keeps it a notch below the top.

DeepL Pro & API

My memory can leave as a TMX export through the API and come back the same way, translated documents come back as files, stored content is deleted 90 days after the agreement ends, and the account can be deleted in the settings. I found no public information on exporting glossaries in TBX or CSV, on a documented migration path from another tool, or on a self-hosted edition, so the exit is solid but not complete.

Google Cloud Translation

Google claims no ownership in the content, translated documents come back through the API or into my own Cloud Storage bucket, and glossaries can be loaded and replaced from TSV, CSV or TMX files with entries listable via the API. We found no documented export of translation memory in an open format, no migration path from another tool, and no stated terms for returning or deleting data when a contract ends.

DeepL Pro & API

DeepL SE is registered in Cologne, the privacy policy states that translation runs exclusively on DeepL's own infrastructure in the EEA, the data protection officer sits in Germany, and ISO 27001, SOC 2 Type II and BSI C5 are attested alongside a published data processing agreement. The captured pages give different indications on processing location — the licence terms also permit third-party cloud infrastructure with DeepL determining the location — and named subprocessors include US firms such as Apryse for the PDF SDK and Cloudflare, so I stop just short of full marks.

Google Cloud Translation

There is an Irish entity with a named managing board in the legal notice, a published Cloud Data Processing Addendum with GDPR commitments, and the Advanced API offers regionalized endpoints configurable for data location under the AI/ML Data Location terms. However, the default global endpoints cannot be configured to keep data in a specific region, the parent company sits in California, and we found no published subprocessor locations or named hosting infrastructure for this service.

DeepL Pro & API

I can compute my yearly cost from public pages alone: Individual at €7.49 per user per month billed annually through Business at €49.99, with per-seat limits on characters, file translations and glossaries spelled out, the Write Pro add-on at €6.99 per user per month, API usage at €22.00 per extra million characters with included volumes published, 20% off annual billing, and VAT treatment stated. Enterprise is custom-priced on request and the fair usage policy behind the unlimited characters is not quantified, which is what keeps it from the top.

Google Cloud Translation

Every model is priced publicly by unit: $20 per million characters for the general model with the first 500,000 characters per month free, $10 per million characters for the LLM model, $25 per million characters for adaptive, volume bands for custom models descending to $30, $0.08 per page and $0.25 per page for document translation, $0.15 and $0.50 per page per target language on the managed Hub, plus one- and three-year discount plans and $300 in new-customer credits under the stated terms "no automatic charges, no commitment". VAT treatment is not stated, and we found no self-hosted edition or its cost on the captured pages.