Translation Cost Per Word in 2026: Human vs Machine vs LLM

How much does translation actually cost? The answer ranges from $0.00001 per word to $0.40 per word depending on the method, language pair, and quality requirements. That's a 40,000x spread, which means the choice of approach matters more than almost any other technical decision in your localization pipeline.

Here are real numbers for 2026.

Human translation rates

Professional human translation rates vary by language pair, subject matter, and translator experience. These are typical rates for English-source translation in 2026:

| Target language | General content | Technical/Legal | Marketing/Creative | | -------------------- | --------------- | --------------- | ------------------ | | Spanish | $0.08-0.12/word | $0.12-0.18/word | $0.15-0.25/word | | French | $0.09-0.14/word | $0.14-0.20/word | $0.16-0.28/word | | German | $0.10-0.15/word | $0.15-0.22/word | $0.18-0.30/word | | Japanese | $0.12-0.18/word | $0.18-0.28/word | $0.22-0.35/word | | Korean | $0.11-0.16/word | $0.16-0.25/word | $0.20-0.32/word | | Chinese (Simplified) | $0.08-0.12/word | $0.12-0.18/word | $0.15-0.25/word | | Arabic | $0.10-0.15/word | $0.15-0.22/word | $0.18-0.28/word | | Portuguese (BR) | $0.07-0.11/word | $0.11-0.16/word | $0.14-0.22/word |

These rates come from freelancer platforms (ProZ, TranslatorsCafe), LSP (Language Service Provider) quotes, and industry surveys. Individual freelancers are at the low end; established LSPs with project management and QA are at the high end.

Turnaround time matters too. A professional translator produces 2,000-3,000 words per day for technical content, 1,500-2,500 for creative content. A 50,000-word documentation set takes a single translator 3-4 weeks.

The hidden costs: project management, review/proofreading (typically an additional 30-50% of translation cost), terminology management, and coordination across multiple translators for large projects.

NMT API pricing

The major NMT (Neural Machine Translation) services in 2026:

| Service | Price per million characters | Approx. per word\* | | ----------------------------- | ---------------------------- | ------------------ | | Google Cloud Translation (v3) | $20 | $0.00010 | | Amazon Translate | $15 | $0.000075 | | Microsoft Translator | $10 | $0.000050 | | DeepL API Pro | $25 | $0.000125 | | DeepL API Free | $0 (500K chars/month) | $0.00 |

\*Assuming average English word length of 5 characters.

Google, Amazon, and Microsoft all offer free tiers (around 500K characters/month). Beyond that, the pricing is per-character with volume discounts.

Real cost example: Translating a 50,000-word documentation set (approximately 250,000 characters) into 5 languages:

• Google Translate: 1.25M characters = $25
• Amazon Translate: 1.25M characters = $18.75
• DeepL: 1.25M characters = $31.25

That same project with human translators: roughly $25,000-$50,000 depending on language pairs and content type. NMT is 1,000x cheaper.

LLM-based translation pricing

LLM translation costs depend on the model, input/output token counts, and how you prompt. Translation typically uses 1.5-2x the input tokens as output (the translation plus the prompt overhead).

| Model | Input cost / 1M tokens | Output cost / 1M tokens | Approx. per word\* | | ---------------- | ---------------------- | ----------------------- | ------------------ | | GPT-4o | $2.50 | $10.00 | $0.0008-0.0015 | | GPT-4o-mini | $0.15 | $0.60 | $0.00005-0.0001 | | Claude Sonnet | $3.00 | $15.00 | $0.001-0.002 | | Claude Haiku | $0.25 | $1.25 | $0.0001-0.0002 | | Gemini 2.0 Flash | $0.10 | $0.40 | $0.00003-0.00006 |

\*Rough estimates including prompt overhead; actual costs vary by language pair (some languages use more tokens per word).

Real cost example: Same 50,000-word documentation set into 5 languages:

• GPT-4o: ~$10-18
• GPT-4o-mini: ~$0.60-1.20
• Claude Sonnet: ~$12-25
• Gemini Flash: ~$0.40-0.75

The smaller/faster models (GPT-4o-mini, Gemini Flash, Claude Haiku) are cost-competitive with NMT APIs while producing better quality for most language pairs. The premium models cost 5-10x more but deliver the best quality.

The comparison table

Putting it all together for a typical project (50,000 words, English to 5 languages):

| Method | Cost | Quality | Turnaround | Best for | | --------------------------- | -------------- | -------------- | ---------- | -------------------------------- | | Human professional | $25,000-50,000 | Excellent | 3-4 weeks | Legal, marketing, creative | | Human + TM tools | $15,000-35,000 | Excellent | 2-3 weeks | Ongoing projects with repetition | | NMT (Google/Amazon) | $18-31 | Acceptable | Minutes | Bulk content, internal use | | LLM (premium models) | $10-25 | Good-Very Good | Minutes | User-facing documentation | | LLM (small models) | $0.40-1.20 | Good | Minutes | High-volume, cost-sensitive | | MTPE\* (NMT + human review) | $5,000-15,000 | Very Good | 1-2 weeks | Best quality/cost tradeoff | | MTPE\* (LLM + human review) | $3,000-10,000 | Excellent | 1-2 weeks | Premium content at lower cost |

\*MTPE = Machine Translation Post-Editing

The MTPE sweet spot

Machine Translation Post-Editing is where a human translator reviews and corrects machine-translated output rather than translating from scratch. This is the approach most companies should be using for professional content.

Post-editing rates are typically 40-60% of full translation rates:

• Light post-edit (fix errors only): $0.03-0.06/word
• Full post-edit (fix errors + improve style): $0.05-0.10/word

The math works out well. For our 50,000-word, 5-language example:

Machine translate with LLM: ~$15

Human post-edit at $0.05/word: $12,500

Total: ~$12,500 vs $25,000-50,000 for pure human translation

You get 80-90% of the quality at 30-50% of the cost, with faster turnaround because post-editing is about 2x faster than translating from scratch.

Token economics for East Asian languages

A critical detail that most cost analyses miss: token counts vary dramatically by language. Chinese, Japanese, and Korean use significantly more tokens per word than European languages in most LLM tokenizers.

A sentence that's 20 tokens in English might be 35-45 tokens in Japanese. This means LLM translation into East Asian languages costs 1.5-2x more than the same content into European languages.

NMT APIs charge per character, which partially compensates (CJK characters encode more meaning per character). But LLM APIs charge per token, and the tokenizer overhead is real.

Budget accordingly. If your primary target languages are Japanese and Korean, increase your LLM translation budget estimates by 50-80% over what European language calculations suggest.

Cost optimization strategies

Translation Memory (TM). Store every translated segment. When the same or similar text appears again, reuse the existing translation instead of paying again. A mature TM can reduce new translation volume by 30-60% for documentation with repetitive content.

Tiered quality. Not all content deserves the same investment:

• Legal/compliance: Human translation, always
• Marketing/landing pages: LLM + human post-edit
• Documentation: LLM translation, human review for top pages
• Internal tools/admin UIs: NMT, no human review
• Log messages/debug output: Don't translate

Caching. If you're using a translation API for dynamic content, cache aggressively. The same UI strings get translated thousands of times if you don't cache. auto18n has built-in caching, but if you're hitting APIs directly, implement your own.

Incremental translation. Only translate what changed. Track content hashes and skip unchanged segments. This is especially important for documentation sites that rebuild frequently but only change a few pages per release.

Batch pricing. Most APIs offer volume discounts. Batch your translation requests rather than sending one-off calls. Some APIs charge less for async/batch endpoints vs synchronous ones.

What to actually budget

For a SaaS company launching in 5 new languages:

| Content type | Volume | Method | Annual cost | | ---------------------- | -------------------- | ------------------------- | ----------------------- | | Marketing site | 20K words | LLM + human review | $2,000-4,000 | | Documentation | 100K words + updates | LLM, human review top 20% | $3,000-6,000 | | App UI strings | 5K strings | LLM | $200-500 | | Blog/content marketing | 50K words/year | LLM + light review | $1,500-3,000 | | Legal/TOS | 10K words | Human translation | $5,000-8,000 | | Total | | | $12,000-22,000/year |

That's roughly the cost of one contractor for one month. For context, the revenue upside of being accessible in 5 new language markets usually dwarfs this cost within the first quarter.

The era where translation was a six-figure line item is over. The question isn't whether you can afford to translate — it's whether you can afford not to.