Best Document Translation APIs in 2026

Document translation APIs are no longer just about sending text to an endpoint. In 2026, developers need to compare accuracy, file format support, latency, pricing, and integration flexibility across providers.

What makes 2026 different is that LLMs now compete directly with dedicated neural machine translation APIs, especially for context-heavy documents, domain-specific terminology, and translation workflows that require reasoning beyond sentence-level text.

This guide compares 12 document translation APIs for teams building translation into applications, data pipelines, internal tools, or AI agents. It looks at how each API handles real document workflows, including PDFs, Word files, spreadsheets, scanned documents, and multilingual content.

Use this table to compare each document translation API by integration type, file format support, language coverage, free tier, pricing, and strongest use case. 

What Is a Document Translation API?

A document translation API translates full files while keeping their structure usable. Instead of sending plain text and getting translated text back, developers send a document such as a PDF, DOCX, PPTX, XLSX, or HTML file. The API extracts the text, translates it, then returns a translated file with the original layout preserved as much as possible.

This matters when formatting is part of the deliverable. Contracts, reports, manuals, invoices, spreadsheets, and slide decks often need tables, headers, footers, footnotes, styles, and page structure to remain intact.

The main decision point is format complexity. Simple text documents are easier to handle. Scanned PDFs, merged spreadsheet cells, embedded images with text, and complex tables require stronger document parsing, OCR, and layout reconstruction.

A general translation API translates text. A document translation API translates the file and rebuilds it into a usable translated document.

‍

Document Translation API vs LLM Translation

Document translation APIs based on NMT, such as DeepL, Google Cloud Translation, Azure Translator, and Amazon Translate, are built specifically for translation workflows. They are usually faster, easier to price at scale, and more reliable for structured documents because many of them support native file formats such as PDF, DOCX, PPTX, XLSX, and HTML.

LLMs such as GPT-4o, Claude, Gemini, and DeepSeek translate through prompting. They do not usually preserve file structure natively, so developers need extra steps for parsing, chunking, formatting, and reconstruction. Their strength is linguistic flexibility. They can handle nuance, tone, intent, terminology instructions, and creative rewrites better than traditional APIs in many cases.

The gap is narrowing. Intento’s 2025 benchmark found that LLMs represented 89% of top-performing translation systems, but they can cost 3-5x more per character. Azure Translator also now offers GPT-4o and GPT-4o mini as model options, which blurs the line between NMT APIs and LLM-based translation.

Use NMT APIs for high-volume structured documents. Use LLMs for shorter, sensitive, creative, or tone-heavy content where wording quality matters more than cost or file preservation.

Top 12 Document Translation APIs in 2026

The best document translators in 2026 are DeepL, Google Cloud Translation, Microsoft Azure Translator, Amazon Translate, SAP Translation Hub, Systran, ModernMT, OpenAI GPT-4o, LibreTranslate, DeepSeek V3, Claude (Anthropic) and Gemini. 

DeepL

DeepL is a quality benchmark for European language pairs. Its API combines NMT with in-house LLMs, making it useful for high-quality output, tone control, and formal writing.

In January 2025, DeepL released its next-generation LLM model, reporting a 1.7x improvement for English to Japanese and English to Chinese, and a 1.4x improvement for English to German. In April 2026, DeepL also made its Voice API generally available, adding real-time speech transcription and speech-to-speech translation. 

DeepL supports PDF, DOCX, PPTX, HTML, 31 languages, plus 81 added as standard in 2025. It includes formality control, glossaries, tag preservation, 500K free chars/month, and $25/M chars Pro pricing.

Pros

• Strong translation quality
• Formality control
• No-credit-card free tier
• Next-generation LLM model

Cons

• Fewer languages than Google
• More expensive than Azure or Amazon
• LLM features still expanding

import deepl

  translator = deepl.Translator("YOUR_API_KEY")

  with open("contract.pdf", "rb") as f:
      translator.translate_document(
          f,
          target_lang="FR",
          output_path="contract_fr.pdf",
          formality="more"  # DeepL-specific: formal register
      )

Google Cloud Translation

Google Cloud Translation is the strongest option for global language coverage, with support for 189 languages. It fits apps that need broad multilingual support and already run on GCP.

Google offers two tiers: Basic v2 for standard text translation, and Advanced v3 for glossaries, batch translation, custom models, and adaptive translation. In 2025, Google added DOC, PPT, and XLS support, improved native PDF handling with shadow text removal, added auto-rotate for scanned PDFs, made adaptive translation generally available, and introduced the translation.googleapis.com endpoint.

Pricing is $20 per million characters, with 500K characters free per month. Document translation is billed separately at $0.08 per page.

Pros

• Widest language coverage
• Better scanned PDF handling
• Adaptive translation
• Strong Google ecosystem integration

Cons

• Billing info required upfront
• Basic vs Advanced can confuse new users
• Document translation is billed per page

Microsoft Azure Translator

Microsoft Azure Translator is one of the best options for cost-sensitive translation workloads. It has the most generous free tier in this comparison, with 2M characters per month and no expiry, plus low pricing at $10 per million characters.

In 2025, Azure added three model options: Azure-MT, GPT-4o, and GPT-4o mini. It also introduced synchronous single-document translation, so developers no longer need Azure Blob Storage for one-off files.

Azure now supports image file translation, translates text embedded in images inside .docx files, and runs on the 2025-10-01-preview API version. It supports 100+ languages, Microsoft 365 and Teams integration, and Custom Translator for domain-specific models.

Pros

• Cheapest per-character pricing
• Best free tier
• GPT-4o model option
• Image translation support

Cons

• Batch document translation still requires Blob Storage
• Behind DeepL for some European pairs
• GPT-4o option priced separately

Amazon Translate

Amazon Translate is a solid choice for teams already building on AWS. It is mainly a text translation API, not a standalone document translation API, so scanned or image-based documents require Amazon Textract before translation.

It supports 75+ languages, real-time translation, batch translation, and Active Custom Translation for domain-specific terminology. Pricing is competitive, with 2M characters per month free for the first 12 months, then $15 per million characters.

The main limitation is product velocity. Amazon Translate had no significant document translation updates in 2025–2026 and now lags behind DeepL, Google, and Azure on file support and LLM options. If you are not already in AWS, there is limited reason to choose it.

Pros

• Strong AWS ecosystem integration
• Competitive pricing
• Active Custom Translation

Cons

• Text-first, Textract required for document workflows
• No GPT/LLM option
• Limited format support vs competitors
• No major updates in 2025–2026

SAP Translation Hub

SAP Translation Hub is an enterprise-grade translation service built for SAP environments and structured business content. It supports formats such as XLIFF, DOCX, and PPTX, making it useful for enterprise localization workflows.

Its main strength is native integration with SAP Business Technology Platform. It also supports translation memory and terminology management, which helps localization teams keep repeated business terms consistent across products, documents, and internal systems.

This is not a general-purpose document translation API. SAP Translation Hub is best for organizations already using SAP or running SAP-based workflows. Pricing is enterprise/custom, and there is no public free tier.

Pros

• Native SAP integration
• Translation memory support
• XLIFF format support

Cons

• Enterprise pricing only
• Limited outside the SAP ecosystem
• Smaller language coverage than Google/Azure

Systran

Systran is one of the oldest machine translation providers, founded in 1968. It is strongest in regulated industries such as defense, legal, and government, where control, deployment model, and data handling matter as much as translation quality.

Its main differentiator is deployment flexibility. Systran offers both a cloud API and on-premise deployment, which is critical for air-gapped, private, or data-sovereign environments.

Systran supports PDF, DOCX, PPTX, and XLSX, covers 50+ languages, and supports custom domain models. Pricing is custom/enterprise, with a trial available. It is best for regulated teams that need secure document translation inside controlled infrastructure.

Pros

• On-premise deployment option
• Domain-specific models
• Strong for regulated or sensitive content

Cons

• No public pricing
• Less consumer-facing than Google or DeepL
• Smaller community and documentation

ModernMT

ModernMT is an adaptive machine translation engine built around translation memory. It learns from previous translations in real time, so output becomes more consistent as teams use it with their own content.

Owned by Translated, ModernMT supports 200+ languages and is designed for professional translators and localization teams. It is especially useful when terminology consistency matters across large projects, product documentation, or recurring customer content.

ModernMT supports DOCX but remains primarily text-focused. It also integrates with CAT tools such as memoQ and SDL Trados. Pricing is custom or usage-based, with trial credits available.

Pros

• Adaptive learning from translation memory
• 200+ supported languages
• CAT tool integrations

Cons

• Limited document format support
• Less known than the big 4 providers
• No fixed public pricing

OpenAI GPT-4

OpenAI GPT-4o is not a dedicated translation API. Translation is done through prompting, which makes it strongest when context, tone, and style matter, such as marketing copy, legal clauses, and creative content.

It has no native document format support. Developers must extract the text, send it to the model, then rebuild the file structure, so it is not ideal for high-volume structured document translation.

GPT-4o supports 90+ languages. Pricing is token-based, around $30–60 per million characters depending on model tier, with no dedicated free tier. LLMs now represent 89% of top performers in Intento’s 2025 quality benchmarks, but with a clear cost premium versus NMT APIs.

Pros

• High quality ceiling for nuanced content
• Strong tone and style control
• Useful for rare language pairs

Cons

• No file format support
• 3–5x more expensive than NMT APIs
• Slower throughput
• Output length can affect cost

LibreTranslate

LibreTranslate is an open-source, self-hosted translation API. It is the only truly free option in this list when deployed on your own infrastructure, with no character limits and no third-party data sharing.

It supports 30 languages and exposes a simple REST API that works with many apps and automation tools. It can run locally, on a private server, or inside offline and air-gapped environments.

The tradeoff is quality and scope. LibreTranslate is significantly behind commercial APIs, supports text only, and requires infrastructure setup. It is best for privacy-first deployments, zero-budget projects, offline use cases, and early prototypes.

Pros

• Completely free when self-hosted
• No data privacy concerns
• No character limits

Cons

• Lower translation quality
• Text-only, no file formats
• Requires infrastructure setup

DeepSeek V3

DeepSeek V3 is a frontier LLM from a Chinese AI lab, released in December 2024. It quickly became a strong option for Chinese↔English translation, especially for teams that want LLM-level quality at a much lower cost than GPT-4o.

Like GPT-4o, translation is done through prompting. DeepSeek V3 accepts text input only, so developers need to extract document content and rebuild formatting separately.

Pricing is around $1–5 per million characters, making it one of the cheapest LLM translation options. It performs strongly on COMET scores for Chinese, Japanese, and Korean pairs, but data residency matters: the model is Chinese-hosted, which can be a blocker for regulated workloads.

Pros

• Lowest-cost LLM option
• Strong for CJK languages
• Competitive quality vs GPT-4o on benchmarks

Cons

• Text-only
• Data residency concerns for regulated industries
• No document format support

Claude (Anthropic)

Claude is an LLM-based option for translation via prompting. Its main strength is preserving tone, register, and coherence across long or complex documents.

It is especially useful for legal, formal, or high-context content where wording matters. Claude can handle paragraph-level consistency well and is strong for language pairs with different syntax, such as EN↔JP and EN↔AR.

Claude has no native document format support, so developers must send extracted text through the API. It supports 50+ languages, with pricing around $15–75 per million characters depending on the model tier. Eden AI supports Claude, so developers can route translation workflows to Claude through Eden AI’s unified API.

Pros

• Strong for tone and register
• Handles long documents well
• Available via Eden AI

Cons

• No file format support
• Higher cost than NMT APIs
• Slower than dedicated translation APIs

Gemini

Gemini is Google DeepMind’s LLM, separate from Google Cloud Translation API. It translates through prompting and adds multimodal capability, meaning it can process text and image inputs in the same workflow.

It supports 100+ languages, has a free tier through the Gemini API, and is competitively priced for LLM translation at around $7–21 per million characters depending on the model tier. Its multimodal input is useful when documents contain screenshots, scanned regions, or embedded visual text.

Gemini does not return translated files with native document formatting. It works best alongside Cloud Translation API in hybrid pipelines where structured document translation and multimodal reasoning are both needed.

Pros

• Multimodal input support
• Competitive LLM pricing
• Google ecosystem integration

Cons

• No document format output
• Quality varies vs DeepL for European pairs
• Overlaps with Cloud Translation API

Document Translation APIs in 2026 Pricing Comparison

Document translation APIs use different pricing models: per-character, per-page, per-token, or custom enterprise pricing. Comparing raw per-character rates can be misleading without checking free tiers, document billing rules, and whether extra parsing or formatting steps are required.

To calculate true cost, start with your expected monthly volume and check whether the free tier is permanent or temporary. For document workflows, compare per-page billing against per-character billing, especially for PDFs and slide decks. Also include post-processing, OCR, retries, and formatting reconstruction, which can add cost outside the translation API itself.

PDF Translation API: What to Look For

Native PDF vs scanned PDF

Start by checking whether your PDFs are native or scanned. Native PDFs contain selectable text, so the API can extract and translate the content directly. Scanned PDFs are images, which means OCR is required before translation. Google, Azure, and DeepL are the strongest options when scanned PDF handling is required.

Layout preservation

Layout is usually the hardest part of PDF translation. Tables, columns, headers, footers, footnotes, and captions can break after translation, even if the API returns a translated file. Test providers with your real PDFs, not clean sample documents.

File size limits

Check file size limits before choosing a provider. Most APIs limit single document uploads to around 5MB to 50MB. For larger PDFs, use batch translation options such as Azure async document translation or Google batch translation.

Output format

Check what format you get back, not only what format you can submit. Some APIs return a translated PDF, while others return DOCX or another editable format. This matters for review, signatures, and document approval workflows.

OCR quality

For scanned PDFs, OCR quality drives translation quality. If OCR misses text, reads columns in the wrong order, or fails on rotated pages, the translation will be wrong too. Amazon Textract plus Amazon Translate is a common AWS pipeline. Google’s 2025 improvements, including auto-rotate and shadow text removal, are also worth testing for image-heavy PDFs.

How to Choose the Right Document Translation API

Start with your file formats

If your workflow includes PDFs, especially scanned PDFs, start with Google Cloud Translation, Azure Translator, or DeepL. If you only translate extracted text, any provider can work. For text-only workflows, LLMs such as GPT-4o, Claude, Gemini, or DeepSeek are worth considering when quality, tone, or context matters.

Match the provider to your language coverage

If you need around 30 languages, most providers are enough. If you need 100+ languages for a global app, Google Cloud Translation or Azure Translator are safer choices. For Chinese, Japanese, and Korean language pairs, compare DeepSeek and Google.

Choose based on translation volume

For low-volume workflows where quality matters more than price, use DeepL or GPT-4o. For high-volume workloads where cost matters, use Azure Translator or Amazon Translate. For free or self-hosted deployments, use LibreTranslate, but expect lower translation quality.

Use your existing cloud ecosystem

If you already run on AWS, Amazon Translate is the easiest fit. If your stack is based on Azure, Microsoft 365, or Teams, use Azure Translator. If you are already on GCP, use Google Cloud Translation. If you have no cloud preference, Eden AI lets you access multiple providers through one API.

Treat sensitive content separately

For regulated, air-gapped, or data-sovereign environments, use Systran because it supports on-premise deployment. For legal, medical, or formal documents where tone and wording matter, Claude or GPT-4o can produce better output than standard NMT, but at a higher cost.

Eden AI is the flexible option when you want to compare, switch, and route between providers without rebuilding your integration each time.

Integrating All Document Translation APIs in One Platform 

Integrating one translation API is simple. Integrating several becomes harder to maintain because each provider uses different API formats, authentication, SDKs, rate limits, error handling, and billing rules.

Eden AI gives developers one API endpoint to access multiple document translation providers. You can switch from DeepL to Google, Azure, Amazon, or an LLM provider by changing one parameter instead of rewriting your integration.

This is useful when you want to choose the best provider per use case:

• Route high-quality European language pairs to DeepL
• Route high-volume, cost-sensitive jobs to Azure
• Route AWS-native workflows to Amazon Translate
• Route scanned PDF or broad language coverage needs to Google
• Route tone-sensitive content to Claude, GPT-4o, Gemini, or DeepSeek

Eden AI also helps you reduce provider risk. If one provider fails, slows down, or hits a rate limit, fallback routing can send the request to another provider. You can also A/B test providers on your own PDFs, contracts, manuals, or support content before committing.

The main benefit is flexibility. Eden AI gives you one integration for both dedicated NMT APIs and LLM-based translation, including Claude, Gemini, GPT-4o, and DeepSeek.

FAQs - Document Translation APIs