Best LLM for translation — quality vs dedicated MT
LLMs can translate with more context and style control than classic MT, but they are often slower and pricier per word. The best LLM for translation is the one that preserves meaning in your language pair at a cost you accept — which might still be “don’t use an LLM.”
English-centric quality index will mislead you. Include Qwen, Gemini, and other multilingual rows in the shortlist, test the actual pair, and compare blended cost at your monthly word count.
Popular frontier models by quality
Live OpenRouter pricing and Arena quality, cached about an hour. Not a static blog table.
| Model | Provider | Blend $ / 1M | Quality | Context |
|---|---|---|---|---|
| Claude Opus 5 | Anthropic | 15.00 | 93.4 | 1,000,000 |
| Claude Fable 5 | Anthropic | 30.00 | 91.1 | 1,000,000 |
| Qwen3.8 27B | Qwen | 1.71 | 89.0 | 1,000,000 |
| Gemini 3.7 Flash | 2.25 | 88.4 | 1,048,576 | |
| DeepSeek V4 Pro 0423 | DeepSeek | 1.12 | 88.2 | 1,024,000 |
When MT wins
High-volume, low-stakes UI strings often belong on a dedicated translation API. LLMs win when you need glossary control, document-level consistency, or “translate this but keep the jokes.” Cost out both at 10M tokens on the model detail pages.
Document-level context
Sentence-level MT misses pronouns and term consistency. An LLM with a glossary in the prompt and enough context to see the whole section can beat MT on marketing pages. That uses input tokens — watch the price.
Language pairs
A model that is excellent at English↔French can be poor at English↔Thai. Build a 50-segment eval per pair. Arena votes will not save you.
Open weights for on-prem translation
If text cannot leave the building, pick an open-weight multilingual model you can host. Use this site’s open-weight cost/quality table as a prior for which family to try first (often Qwen-class for many Asian languages).
QA loop
Always back-translate or use a second model as a critic for published content. The extra tokens are cheaper than a mistranslated legal page.
FAQ
Is GPT the best translator?
- Not uniformly. Test your language pair. Multilingual specialists and Gemini/Qwen rows often compete.
Are LLMs cheaper than Google Translate?
- Usually not at huge volume. Compare $/million tokens to MT APIs for your word count.
Can I use a cheap model for translation?
- For internal gist, yes. For customer-facing copy, spend time on evals before you spend less on tokens.
Does context window matter?
- For documents, yes. For isolated strings, no.
Related guides
Also try: LLM leaderboard, compare tool, live arena.