DeepL Pro & API vs Google Cloud Translation

DeepL Pro & API

DeepL's own engine covers 126 published languages with a per-language matrix for glossaries, formality, translation memory and style rules, and document translation preserves layout for Word, PowerPoint, Excel, PDF, InDesign and image files. Glossaries adapt terminology grammatically, support priority ordering and non-translated terms, a generator can build them from past translation files, and an account-scoped adaptive capability is stated never to train models. We found no public information on automatic quality estimation per segment or a published translation-quality evaluation with its method; the blind-test superiority claim appears without one.

Google Cloud Translation

The engine side is strong: 189 languages with published language lists, glossaries in TSV, CSV or TMX applied during translation with case handling, an own NMT plus a Gemini-based Translation LLM and adaptive translation that uses the customer's own small datasets, custom AutoML models, and document translation for Word, PowerPoint, PDF and Excel that preserves the original layout. We found no published quality evaluation with its method, no per-segment quality estimation that routes weak output to review, and no public information on a formality or tone setting; machine translation quality prediction scores are only named on the Translation Hub pricing line.

DeepL Pro & API

A translation memory can be uploaded as TMX, shared with the team, activated in the translator and reused on similar texts, and glossaries are managed centrally with sharing and merging. We found no public information on review or approval stages, automated QA checks, comments and screenshots, per-language roles, or inviting translators and agencies inside the product; the CAT-tool connection runs through the paid API and workflow building is described only in outline.

Google Cloud Translation

This is a translation engine rather than a managed workflow: glossaries are kept centrally with create, update, list and delete over the API, multiple glossaries per project, and Translation Hub's pricing line names translation memory, human review and quality-prediction scores. We found no public information on review or approval stages, automated QA checks, roles per language, an in-context editor, or inviting translators and agencies.

DeepL Pro & API

The REST API is documented in detail, with the authentication scheme, status codes, a 1 GB translation-memory import ceiling and a deprecated legacy languages endpoint, and covers document formats including Word, PowerPoint, Excel, HTML, SRT, TXT, PDF, JPG, PNG, InDesign and Contentful JSON. Browser extensions for Chrome, Edge and Firefox, Microsoft 365 and Google Workspace add-ins, Google Drive, SharePoint, Adobe Experience Manager and Zendesk connectors, AWS Marketplace and an MCP integration for ChatGPT, Claude and Cursor are all named. We found no public information on a CLI, SDKs, over-the-air delivery, Git synchronization, CI pipelines or webhooks.

Google Cloud Translation

There is a documented REST API with client libraries for both editions, glossary endpoints, document formats from DOC to XLSX, batch requests reading from and writing to the customer's Cloud Storage, and documented quotas and limits. We found no public information on Git repository sync, CI pipelines, CMS or design-tool connectors, webhooks, a CLI or over-the-air delivery on the captured pages.

DeepL Pro & API

For Pro, texts are stated never to be used for model training and to be deleted after the agreed services, with debug storage capped at 72 hours, requested stored content deleted 90 days after the agreement ends, and TLS encryption, BYOK, SSO over SAML and OIDC, MFA, role-based permissions and audit logs documented; ISO 27001, SOC 2 Type II with its scope and BSI C5 Type 2 are stated and a data processing agreement is an integral part of the terms. Two reservations: the free consumer service processes texts to train neural networks and prohibits personal data there, and the free API Developer tier reserves the right to perpetually store content. We found no public information on a dedicated or private deployment option.

Google Cloud Translation

The data-usage statements are unusually specific and are what I would put before the board: sent text is held briefly in-memory to perform the translation and return the result, is not used to train or improve translation features, is not shared with any third party or made public, no ownership is claimed in the content, glossaries are encrypted under fine-grained IAM permissions, and the Cloud Data Processing Addendum applies. The only AI provider in the chain is Google's own Gemini, so there is no third-party engine chain to restrict; we found no public information on audit logs, single sign-on, certifications with their scope, customer-managed keys or a dedicated deployment option.

DeepL Pro & API

The translation memory exports as TMX through an asynchronous API job and imports up to 1 GB, accounts can be deleted self-service, and requested stored content is deleted 90 days after the agreement ends. We found no public information on glossary export in TBX or CSV, bulk export of projects and history, a documented migration path from other tools, or a self-hosted edition.

Google Cloud Translation

Glossaries import in TSV, CSV or TMX and every entry can be listed and retrieved through the API, and document translation outputs land in the customer's own Cloud Storage bucket. We found no public information on translation-memory export formats, a documented migration path, return or deletion of data on termination, or a self-hosted edition.

DeepL Pro & API

DeepL SE is registered in Cologne with a German VAT ID and an external data protection officer, and the privacy policy states document translation is processed exclusively on DeepL's infrastructure in the EEA, while the terms reserve processing on third-party cloud infrastructure with DeepL determining the location, so the captured pages give different emphases on where processing sits. Subprocessors are named, Cloudflare in San Francisco for CDN, Stripe Payments Europe in Dublin and Payrails in Berlin for payment, ElevenLabs for backup voices and Apryse Systems for the PDF SDK, the last two without a stated location, and the translation engine itself is DeepL's own. A data processing agreement is offered and ISO 27001, SOC 2 Type II with scope and BSI C5 Type 2 are published; we found no public information on a self-hosted edition.

Google Cloud Translation

EU customers contract through Google Cloud EMEA Limited in Dublin, Google Ireland Limited appears as service provider in the legal notice with an Irish registration number, and a Cloud Data Processing Addendum is published and incorporated into the agreement. Against that, the Basic API and the default endpoints are global and data cannot be configured to remain in a specific region — only the Advanced API offers regionalized endpoints configurable for data location — the parent is Google LLC in California, and we found no public information on hosting locations or the subprocessor list from the captured facts.

DeepL Pro & API

Seat prices are public, €7.49, €24.99 and €49.99 per user/month billed annually for Individual, Team and Business, with each tier's character, file-translation and glossary limits, and the API tiers publish €23.80 per month billed annually plus €22.00 per 1,000,000 extra characters, €1.96 per extra speech-to-text hour and €3.93 per extra speech-to-speech hour, the Write Pro add-on at €6.99 per user/month billed annually, VAT treatment, the annual-billing saving of 20 % and accepted payment methods. The captured pages give different signals on SSO, listed as a Team feature in one capture and contact sales in another, Enterprise pricing is custom, and the fair-usage policy behind unlimited characters is not detailed. We found no public information on the minimum contract term or the price of the File Translation Add-On.

Google Cloud Translation

Usage pricing is published to the cent with its scaling rules: first 500,000 characters per month free then $20 per million characters for NMT (with one- and three-year rates of $18.00 and $16.00), $10 per million input and output characters for the Translation LLM, $25 per million for adaptive translation, document translation from $0.08 per page, Translation Hub from $0.15 per page per target language, custom-model training at $45 per hour capped at $300 per job, volume bands for custom translation down to $30.00, $300 in free credits for new customers, and a stated basis of no automatic charges and no commitment. What remains unstated is VAT treatment and any support or setup costs.