DeepL Pro & API vs Google Cloud Translation

DeepL Pro & API

Their own engine covers 126 published languages with per-language feature support queryable through the languages endpoint, glossaries with grammatical adaptation and do-not-translate terms, a formality setting, style rules and custom instructions, and document translation that keeps formatting for Word, PowerPoint, PDF and even InDesign. The glossary generator digests your own sentence pairs and adaptive translation stays scoped to your account, which edges toward real customisation, but quality is only claimed through unnamed blind tests and I found no public information on per-segment quality estimation or a published evaluation method.

Google Cloud Translation

This is the engine itself and it is a serious one: 189 languages with published lists, glossaries in TSV, CSV and TMX applied during translation, document translation that preserves formatting for Word, PowerPoint, PDF and Google Workspace, plus Adaptive Translation that works from my own small datasets without a training job and AutoML custom models on 53 language pairs to English. What keeps it off the top band: I found no published quality evaluation with its method, no formality or tone setting, and quality prediction scores appear only in the Translation Hub advanced tier rather than as per-segment routing of weak output to human review.

DeepL Pro & API

Translation memory is real and API-driven — TMX upload, sharing, activation in the translator, and reuse of past translations of similar texts — and glossaries can be merged, prioritised across teams and created with forbidden terms. The review side is where an ongoing localisation process lives, though, and I found no public information on review or approval stages, automated placeholder or terminology checks, roles per language, or inviting translators and agencies into the product. A "Translation Flow" capability is named on the translator page without workflow detail, and connecting CAT tools is supported on the paid API for teams needing a fuller review process elsewhere.

Google Cloud Translation

The management layer is thin for someone living in JSON and xcstrings: glossaries get real central handling — create, update, delete, entry-level operations, IAM permissions, multiple glossaries per project — and Translation Hub's advanced tier lists translation memory, human review and quality prediction scores. But the pages state plainly that glossaries have no version control, and I found no public information on automated QA checks for placeholders or tags, comments and screenshots, an in-context editor, review stages, or roles that limit translators to their languages.

DeepL Pro & API

The REST API is properly documented and versioned — translation memory import and export are described down to status codes, signed upload URLs and a gigabyte size cap — and the connector list is broad: Microsoft 365, Google Workspace, Adobe Experience Manager, Zendesk, SharePoint, Google Drive, an AWS Marketplace listing, browser extensions for Chrome, Edge and Firefox, and an MCP integration for ChatGPT, Claude and Cursor. For a developer like me the gap is the formats and the pipeline: everything documented is document-oriented (Word, PowerPoint, PDF, HTML, SRT, Excel, InDesign, Contentful JSON), and I found no public information on software localisation formats like XLIFF, PO, JSON keys, YAML, Android XML or iOS strings files, nor on a CLI, SDKs, Git-host sync, webhooks or over-the-air delivery.

Google Cloud Translation

The API surface is solid — documented REST endpoints including full glossary operations, client libraries, published quotas and limits, batch jobs reading from and writing to Cloud Storage buckets, release notes and troubleshooting docs. But for an app developer the gap is glaring: the documented document formats are DOC/DOCX, PDF, PPT/PPTX and XLS/XLSX only, and I found no public information on JSON, iOS strings or xcstrings, Android XML, XLIFF or PO handling, nor on a CLI, Git repository sync with pull requests, over-the-air updates, webhooks or CMS connectors.

DeepL Pro & API

For Pro subscribers the pages state texts are never used to train DeepL's models and are deleted after the service is performed, with storage only as technically needed, a 72-hour encrypted debug exception, and customer-requested stored content deleted 90 days after the agreement ends; TLS and advanced encryption, bring-your-own-key encryption, SSO over SAML and OpenID Connect, audit logs and role-based permissions are all stated. ISO 27001, SOC 2 Type II with a stated scope, BSI C5 Type 2 and HIPAA compliance are listed, a data processing agreement sits in the Trust Centre, and the audit verdict is downloadable. Caveats: the free tier does train on uploaded content and the free API tier reserves the right to store content perpetually, and I found no public information on a private or dedicated deployment option.

Google Cloud Translation

The data-usage documentation is unusually concrete: sent text is held briefly in-memory to perform the translation and is not stored on servers, Google states it does not use content to train or improve its translation features, will not share it with third parties or make it public, claims no ownership, and keeps glossaries encrypted under fine-grained IAM in my own project, with a published Cloud Data Processing Addendum. What is missing for the top band: I found no public information on audit logs, certification scope, or a dedicated deployment option in the captured pages.

DeepL Pro & API

The translation memory can leave: export runs as a background job producing a TMX file (stated as the only export format), import takes TMX up to 1 GB through a signed-upload flow, and content stored at the customer's request is deleted 90 days after the agreement ends. Glossaries can be created from file uploads and a generator that digests TMX or TSV, but I found no public information on exporting a glossary in an open format such as TBX or CSV, on a documented migration path from named competitors, or on any self-hosted or open-source edition.

Google Cloud Translation

The raw assets stay mine: translated documents land in my own Cloud Storage bucket, glossaries import from TSV, CSV or TMX and every entry is readable through the API, and Google claims no ownership in transmitted content. But there is no documented translation memory export, no complete project export in open formats, no migration path from another tool, and I found no public information on a self-hosted or open-source edition or on termination terms.

DeepL Pro & API

The contracting entity is DeepL SE in Cologne with a German commercial register entry and VAT ID, the privacy policy states document translation is processed exclusively by DeepL on its own infrastructure in the EEA, and the machine translation engine is DeepL's own, so no foreign MT provider sits in the chain. Subprocessors are named with addresses — Cloudflare in San Francisco for CDN, Stripe in Dublin and Payrails in Berlin for payments, Usercentrics in Munich for consent, Apryse for the PDF SDK — alongside ISO 27001, SOC 2 Type II and BSI C5 stated with the data processing agreement published. One caveat: the Pro licence terms reserve processing on third-party cloud infrastructure with the location determined by DeepL, so the captured pages give different statements on where content may run.

Google Cloud Translation

The contracting side looks European: Google Ireland Limited is registered in Dublin with a published register number and VAT ID, and a Cloud Data Processing Addendum is published and incorporated into the agreement. But the captured pages do not confirm where texts are hosted — global endpoints mean data "cannot be configured to remain in a specific region for processing", with only the Advanced API offering regionalized endpoints for data location — and although a subprocessor list page was captured, it yields no named subprocessors here, while the parent is Google LLC in California.

DeepL Pro & API

Seat prices are published — €7.49, €24.99 and €49.99 per user per month billed annually for the translator tiers, Enterprise on request — and the API tiers are unusually computable: a free Developer tier with a one-time credit of 1 million characters, Growth at €23.80 per month billed annually plus €22.00 per 1,000,000 extra characters, with VAT treatment and a 20% annual discount stated. What keeps me from fully budgeting: "unlimited" characters sit under a fair usage policy whose bounds are not quantified, the File Translation Add-On appears without a price, the SSO page says contact sales even though SSO is listed in the Team plan, and I found no public information on a minimum term.

Google Cloud Translation

I can genuinely compute the annual bill from public pages: text translation at $20 per million characters above a free 500,000 per month, document translation at $0.08 per page and $0.25 per page, Translation Hub at $0.15 and $0.50 per page per target language, adaptive and LLM translation at $25 and $10 per million characters, custom-model volume bands down to $30.00, training at $45 per hour capped at $300 per job, one- and three-year committed-use prices, $300 in new-customer credits, and "No automatic charges, no commitment." What keeps it from the top: VAT treatment beyond the USD note is not stated, and cancellation and data-return terms are not on the pricing pages.