DeepL API vs Google Cloud Translation: budget for file handoff and minimum usage in batch manual translation

Prioritize DeepL API for uploading individual manuals and retrieving editable files, or Google Advanced when Cloud Storage and multifile, multilingual batches are already needed. Glossary direction, PDF type and per-file minimum billing can change the choice.

Source: docs.cloud.google.com
November 9, 2026
5 min

For teams translating batches of product manuals, DeepL API fits a workflow that sends individual files from an existing file system and retrieves editable results. Teams already managing files in Google Cloud and needing multiple files and target languages in one job can prioritize Cloud Translation Advanced. The decisive issues are file handoff, whether terminology is applied in the correct language direction, and how short files are billed. A few translated passages cannot establish which whole workflow fits better.

Start with the right document API

This comparison concerns developer APIs, not DeepL website translation plans versus the Google Translate website. Google's format support table places synchronous and batch document translation in Advanced (v3), using NMT or AutoML Translation; Basic (v2) cannot replace that workflow. The availability of Translation LLM in Advanced does not justify applying its text-model capabilities or prices directly to the document API.

DeepL's document API supports DOCX and PDF, as well as PPTX, XLSX, HTML, TXT and other formats. Its current usage limits allow 100 MB and 5 million characters per Word/PDF file on paid API plans; the free API limits are 10 MB and 500,000 characters. These are file limits, not website-plan allowances that can be transferred to the API.

File handoff determines who maintains the job queue

DeepL's official example uses three steps: upload, poll and download. After uploading, retain the document_id and document_key, then download the result when ready. A batch of manuals means managing jobs and saved-file status for every file and target language. Server-side files are deleted after the first successful download, so save the result reliably before distributing it to reviewers.

Google's batch API reads files from Cloud Storage and writes results to the output location, returning a long-running operation whose status can be checked; a request can specify up to 10 target languages. It requires a source language, and the glossary and API call must be in the same region. The output index.csv separately records ordinary translations, glossary-applied translations and errors. The review system should receive the glossary version rather than declare completion after finding the first output file. Teams already managing cloud storage and permissions can accommodate this workflow more readily; otherwise, that infrastructure adds integration work.

A glossary also needs the correct language direction

DeepL's glossary management guide defines each dictionary as a one-way language pair. An English-to-German dictionary does not automatically cover German-to-English; the reverse direction needs another dictionary. Document translation with a glossary still requires source_lang. Do not transfer the text REST API's automatic source-language detection behavior to documents. Also check the document API's language capabilities to confirm that both source and target support glossary.

Google supports unidirectional glossaries and multilingual equivalent term sets. Matching is case-sensitive by default, and some stopwords cause glossary entries to be ignored. Its official guide also advises retaining original glossary files because glossary resources lack version control. Whichever service you choose, associate each manual delivery with the glossary version and language direction used, so terminology changes in the next revision can be explained.

Editable output does not eliminate layout rework

DeepL can request PDF-to-DOCX conversion through output_format. Google's document guide restricts PDF-to-DOCX conversion to batch translation of native PDFs; scanned PDFs are outside that conversion scope. Google also states that scanned content in PDFs combining native text and scans is not translated, nor is content inside DOC/DOCX text boxes. If a manual contains such material, these boundaries can change the service choice or preprocessing plan.

When DOCX source files are available, prioritize evaluating those before generating delivery PDFs. When only PDFs are available, first distinguish native, scanned and mixed files. Scan recognition, complex tables and multicolumn layouts need separate checks. These documented capabilities do not establish that either service preserves your manuals' layouts better, nor do they establish a translation-quality winner.

Budget for the distribution of files

DeepL's API billing guide specifies a minimum of 50,000 source characters per translation of files such as DOCX and PDF; larger files are counted at their actual character count. Suppose 100 short manuals each contain only 5,000 characters and are each translated into 2 languages. Billable usage is 100 × 2 × 50,000 = 10,000,000 characters, rather than the 1,000,000 obtained by multiplying actual source characters by target languages. Apply the unit rate and fixed fees of the applicable API contract to this billable volume to build a comparable budget.

Google's published standard pricing lists NMT DOCX/PDF document translation at USD 0.08 per page and AutoML custom models at USD 0.25 per page; target languages multiply billable pages. If the same batch totals 400 billable pages and is translated into 2 languages, NMT translation costs 400 × 2 × 0.08 = USD 64. This hypothetical calculation excludes other services such as cloud storage and does not use committed-spend discounts. XLSX appears under character-based billing, so page pricing cannot be applied to every format.

Consequently, when there are many short DOCX/PDF files, first calculate DeepL's per-file minimum usage and compare it with Google's actual billable pages. For longer files, recalculate characters, pages and human rework costs. If the existing system suits individual uploads and downloads, DeepL merits the first trial. If multilingual batch processing and Cloud Storage are already everyday infrastructure, Google Advanced aligns more closely with the existing handoff. Ultimately, use the same set of manuals approved for evaluation to check terminology matches, untranslated areas, editability and actual invoices before expanding the volume.

Content note:This article is based on publicly available information and is intended as a reference for evaluating tools. Features, pricing, and related policies may change at any time; please refer to the latest official information.

Original source:docs.cloud.google.com