Machine Translation APIs: DeepL, Google Translate, AWS Translate, Azure Translator, OpenAI / Claude, Lilt, ModernMT

⬅️ Backend & Data Overview

If you're building a B2B SaaS in 2026 that needs real-time text translation — translating user-generated content, AI chat responses, in-product strings on demand, customer support tickets across languages — you need a machine translation (MT) API. The naive approach: Google Translate widget. The structured approach: pick an API based on quality (DeepL leads for European languages; Google for breadth; LLM-based for context), price per character, language coverage, and whether you need glossary support / formality control / domain customization. The right pick depends on volume, language pairs, and whether you also need a TMS (separate concern; see localization-translation-tools.md).

TL;DR Decision Matrix

The first decision is quality vs price vs language coverage: DeepL leads quality for major European languages; Google leads coverage (133+ languages); LLMs lead context-awareness; OSS leads privacy. The second decision is integration scope: just translation API vs full TMS workflow.

Decide What You Need First

Real-time chat / message translation (the 30% case)

You need to translate user messages between languages on the fly (chat apps, customer support, social).

Right tools:

• Google Translate — broad language coverage
• DeepL — better quality for EU languages
• Azure Translator — cost-effective
• LLM-based (Claude / GPT-4o) — context + nuance

AI-generated content translation (the 25% case)

Your AI feature generates content; need it in multiple languages.

Right tools:

• LLM-based (Claude / GPT-4o) — generate in target language directly
• DeepL / Google — translate after generation

Customer support across languages (the 20% case)

Support tickets in dozens of languages; agents speak few.

Right tools:

• DeepL — quality matters
• Google Translate — breadth
• Azure Translator — bundled with Microsoft helpdesk products

Document translation (the 15% case)

Translate uploaded PDFs / DOCX / XLSX files.

Right tools:

• DeepL Pro — supports DOCX, PPTX, PDF directly
• Google Translate API — text only; pre-process docs
• Smartcat — hybrid MT + human review
• Lilt — enterprise post-edit

High-quality / branded translation (the 10% case)

Marketing content; legal docs; product descriptions where quality matters.

Right tools:

• DeepL with glossary
• Lilt with human-in-loop
• ModernMT with domain training
• Claude / GPT-4o with prompted style

Provider Deep-Dives

DeepL — quality leader for European languages

Founded 2017 (Germany). Considered best-quality MT for major languages.

Pricing in 2026:

• Free: 500K chars/mo
• API Free: 500K chars/mo (separate)
• Pro: $9-25/user/mo (browser/web)
• API Pro: $5.50/1M chars + $0.000005/char beyond
• Enterprise: custom

Languages: 33+ (English, German, French, Spanish, Portuguese, Italian, Dutch, Polish, Russian, Chinese, Japanese, Korean, etc.)

Features: high-quality translation; glossary support; formality control (formal/informal); document translation (DOCX, PDF, PPTX); CAT-tool integration.

Why DeepL wins: quality measurably better than Google for major European languages; glossary + formality unique among MT APIs.

Trade-offs: smaller language list (33 vs Google's 133+); pricier than Google.

Pick if: quality matters; major European languages; brand-conscious. Don't pick if: need many low-resource languages.

Google Translate API — breadth leader

Pricing in 2026: 500K chars/mo free; $20/1M chars (Cloud Translation Basic); Advanced (custom models) $80/1M chars.

Languages: 133+ (most of any commercial provider).

Features: text translation; document translation; auto-detect; custom models (with training); glossary; batch translation.

Why Google: broadest language support; reliable; well-documented; familiar.

Trade-offs: quality good but not best for major languages; no formality control (formality limited).

Pick if: need broad language coverage. Don't pick if: quality-first for European languages (DeepL stronger).

AWS Translate — AWS-native

Amazon's MT service.

Pricing in 2026: $15/1M chars; 2M chars/mo free for first 12 months.

Languages: 75+.

Features: real-time + batch; custom terminology; profanity filter; formality (some langs); active custom translation (with human-edited training data).

Why AWS: native AWS integration (Lambda, SDK, IAM); reasonable pricing.

Pick if: AWS-native infrastructure. Don't pick if: quality-first or non-AWS.

Azure Translator — Microsoft-native

Microsoft's MT service.

Pricing in 2026: 2M chars/mo free; $10/1M chars S1 (cheap); custom $40/1M chars.

Languages: 100+.

Features: real-time + batch; document translation; custom translator (train on your data); transliteration; dictionary lookup; bilingual dictionary; profanity filter.

Why Azure: cheapest enterprise MT; broad language; custom training; Microsoft 365 integration.

Trade-offs: less brand recognition; quality similar to Google.

Pick if: Azure-native; cost-priority; need custom training. Don't pick if: don't trust Microsoft.

LLM-based translation (Claude / GPT-4o)

Use vision / chat LLMs for translation.

Pricing in 2026:

• Claude Sonnet 4.6: $3/1M input + $15/1M output tokens
• Claude Opus 4.7: $15/1M input + $75/1M output tokens
• GPT-4o: $2.50/1M input + $10/1M output

Why LLM-based:

• Context-aware (knows surrounding sentences)
• Nuanced (idioms, tone, register)
• Customizable via prompts ("translate informally", "preserve legal terms")
• Long-context (entire docs in single call)
• Good for low-resource languages improving rapidly

Trade-offs:

• Cost per request varies (1 character ≈ 0.25 tokens roughly)
• Slower than purpose-built MT
• Hallucination risk (rare but possible)
• Rate limits

Pick if: low-medium volume; nuance matters; need style control. Don't pick if: high volume + cost-sensitive.

Lilt — adaptive MT with human-in-loop

Founded 2015. Enterprise MT + post-edit.

Pricing: enterprise custom.

Features: adaptive MT (learns from corrections); human-in-loop post-edit; CAT tool; project management.

Pick if: enterprise; high-stakes content; willing to invest in human review. Don't pick if: pure-API SMB.

ModernMT — adaptive domain MT

Adaptive MT engine.

Pricing: $30+/1M chars enterprise.

Features: domain adaptation; segment-level training; integration with TMS.

Pick if: domain-specific content (legal, medical). Don't pick if: general-purpose.

Yandex / Baidu — regional specialists

Strong in Russian (Yandex) and Chinese (Baidu).

Pick if: heavy Russian or Chinese workflow.

LibreTranslate / Argos Translate — OSS

Self-hosted OSS MT.

Pricing: free; you host.

Quality: 60-80% of commercial; varies by language.

Pick if: privacy-critical; offline; cost-priority.

What MT APIs Won't Do

Buying an MT API doesn't:

1. Replace human translators for high-stakes content. Legal contracts, marketing campaigns benefit from human review.
2. Handle context across documents. API translates the input string; doesn't know related context unless you provide it.
3. Solve cultural localization. Translation ≠ localization. Date formats, currencies, holidays, measurement units all need separate handling.
4. Preserve brand voice automatically. Glossary + style instructions help; not perfect.
5. Translate dynamic content perfectly. "Click here" and "here" link text break in non-English flow.

The honest framing: MT is 80-95% of the way there for most content. The last 5-20% requires human or LLM nuance.

Pricing Comparison — Realistic Volume

For 10M chars/month (e.g., chat translation in B2B SaaS):

• DeepL API Pro: 10M × $5.50/1M = $55/mo
• Google Translate: 10M × $20/1M = $200/mo
• AWS Translate: 10M × $15/1M = $150/mo
• Azure Translator (S1): 10M × $10/1M = $100/mo
• Claude Sonnet 4.6: ~2.5M tokens × $3 input + 2.5M × $15 output = ~$45/mo (varies)
• GPT-4o: ~$30/mo at similar volume

For 100M chars/month:

• DeepL Enterprise: negotiate; typically $300-500/mo
• Google Translate: $2,000/mo
• Azure (S1): $1,000/mo
• Claude Sonnet: ~$450/mo
• Self-hosted LibreTranslate: hosting only ($50-200/mo)

The cost surprise: LLM-based translation is competitive at low-medium volume; specialty MT cheaper at high volume.

LLM vs Traditional MT — 2026 Decision

Decide LLM-based vs traditional MT.

Use LLM-based (Claude / GPT-4o) when:
- Context matters (surrounding sentences inform translation)
- Need style control (formal / casual / brand voice)
- Low-medium volume (<50M chars/mo)
- Nuance over speed
- Multi-language at once (one call, many outputs)

Use traditional MT (DeepL / Google / Azure) when:
- High volume (>100M chars/mo)
- Speed-priority (LLMs slower)
- Specific language quality (DeepL for EU)
- Custom training (Azure Custom Translator, Google AutoML)
- Compliance requires deterministic output

Use OSS (LibreTranslate) when:
- Privacy-critical
- Cost-priority
- Offline / air-gapped

Hybrid pattern (recommended for many):
- Tier 1: cheap MT for bulk (Azure / Google)
- Tier 2: LLM for quality-critical
- Tier 3: human for highest-stakes

For [USE CASE]:
1. Recommended approach
2. Cost calculation
3. Quality tradeoffs
4. Latency expectations
5. Migration path if needs change

The 2026 trend: LLMs are competitive with specialty MT for many use cases. DeepL still wins for major European-language quality + cost; LLMs win for context + style control.

Pragmatic Stack Patterns

Pattern 1: Indie SaaS adding MT ($0-50/mo)

• DeepL Free OR Azure Translator Free Tier
• Or LLM-based at low volume
• Total: free initially

Pattern 2: SMB B2B SaaS ($50-500/mo)

• DeepL API Pro for quality
• Or Azure Translator for breadth + cost
• 10-100M chars/mo

Pattern 3: Mid-market ($500-2K/mo)

• DeepL Pro or negotiated enterprise
• Glossary + formality control
• Custom terminology

Pattern 4: AI-product translation ($cheap-expensive)

• LLM-based via Vercel AI Gateway
• Context-aware translation
• Multi-step: generate → translate or generate-in-target-language

Pattern 5: Enterprise legal / brand ($$$+)

• Lilt or Smartling with human-in-loop
• Translation memory
• Post-edit workflow

Pattern 6: Privacy / OSS ($hosting)

• LibreTranslate self-hosted
• Or specialty open-source models (NLLB, Marian)

Pattern 7: Document translation

• DeepL Pro for DOCX / PDF native
• Or convert to text first → translate → re-format

Decision Framework: Three Questions

1. What's the use case?

   • Chat / messages → Google / DeepL / Azure / LLM
   • AI content → LLM-based
   • Customer support → DeepL / Google
   • Documents → DeepL Pro / Smartcat
   • High-stakes → Lilt / Smartling + human review

2. What languages matter?

   • Major EU languages → DeepL (quality)
   • Broad coverage → Google (133+)
   • Russian → Yandex
   • Chinese → Baidu
   • Many low-resource → Google / LLM-based

3. What's your scale + budget?

   • <500K chars/mo → free tiers
   • 500K-50M chars/mo → DeepL / Azure
   • 50M-500M → Azure / Google + negotiate
   • 500M+ → enterprise contracts; consider self-host

Verdict

For 30% of B2B SaaS in 2026 needing MT: DeepL for European-language quality.

For 25%: Google Translate for breadth.

For 15%: Azure Translator for cost + Microsoft alignment.

For 15%: LLM-based (Claude / GPT-4o) for context + style.

For 5%: Lilt / Smartling for enterprise / human-in-loop.

For 5%: Yandex / Baidu for Russian / Chinese specialty.

For 5%: LibreTranslate for OSS / privacy.

The mistake to avoid: defaulting to Google because of name recognition. DeepL is measurably better for major EU languages at similar price. Test both on actual content.

The second mistake: using LLMs at high volume without cost analysis. $30/mo at 10M chars vs $300/mo at 100M chars — math first.

The third mistake: expecting MT to handle localization. Translation is words; localization is dates, currencies, formats, cultural references. Different problems.

See Also

• Localization & Translation Tools — TMS for product localization
• Voice AI Providers — TTS / STT
• Email Providers — translated transactional emails
• API Integration — generic patterns
• VibeReference: Anthropic Claude — Claude-based translation
• VibeReference: OpenAI GPT — GPT-based translation
• VibeReference: AI Gateways — unified routing
• VibeReference: Vercel AI Gateway — Vercel AI Gateway
• VibeReference: Document Parsing & OCR Services — extract text before translating
• VibeReference: SMS & Voice APIs — SMS in multiple languages
• VibeWeek: Internationalization — i18n implementation
• VibeWeek: Customer Support — multi-language support
• VibeWeek: AI Features Implementation — AI translation features
• LaunchWeek: International Expansion Playbook — translation in expansion
• LaunchWeek: Vertical SaaS Positioning — vertical-specific terminology