How do I check a translation in a language I cannot read?

Back-translate it on a different model and compare with your original: what survives the round trip is solid, and what changed is where to look. Parallel tabs run both sides on one prompt.

Can it keep my terminology consistent?

Yes. Keep the glossary in a Drive file the chat references, or give a translation agent memory so approved terms persist across jobs instead of being re-pasted.

Formal or informal register: how do I control it?

State it explicitly (audience, formality level, and the pronoun form where the language distinguishes them) and the flagship models hold it; the register keeper picks hold it longest.

All roundups

Best of

Translation that survives a back-check.

Multilingual picks plus the workflow that catches drift: translate on one model, back-translate on another, keep what agrees.

Last reviewed July 16, 2026

Try idapt free Browse all models

Modern flagships translate with context, register, and idiom that phrase-based tools miss, but no public benchmark settles which is best for your language pair. The reliable upgrade is the workflow: translate on one model, back-translate on a second, and reconcile the differences. Every pick below runs in the same composer, which makes that triangulation one prompt away.

How we pickedThere is no named translation benchmark in the live rankings, and this page does not invent one. Picks lean on documented multilingual coverage, general benchmark standing, and consistent results from cross-model verification runs; the back-translation workflow above replaces trust in any single pick. Prices render live beside each.

The picks

1Gemini 3.1 ProThe wide-coverage workhorse.

$2 in / $12 out per M tokens · 1.0M context

Strong across major and less-common pairs, with a context window that holds an entire document and its glossary at once, so terminology stays consistent to the last page.

Best for: Long documents and less-common pairs.

2Claude Fable 5The register keeper.

$10 in / $50 out per M tokens · 1M context

The writing lane's pick applied to translation: it carries tone, idiom, and formality across languages instead of flattening them, which shows in marketing copy and literary text.

Best for: Copy where tone is the point.

3GPT-5.6 SolThe constraint follower.

$5 in / $30 out per M tokens · 1.1M context

Reliable instruction-following at frontier level: hand it a glossary, banned terms, and a style rule and it applies them consistently across a long job.

Best for: Terminology-constrained business translation.

4Qwen3.7 MaxThe Chinese-English flagship.

$1.25 in / $3.75 out per M tokens · 1M context

The open flagship from a lab that documents first-class Chinese-English performance; the natural pick for CJK pairs and the verification partner for them.

Best for: Chinese and other Asian-language pairs.

5Mistral Small 4European languages at volume prices.

$0.15 in / $0.6 out per M tokens · 262K context

Open weights and efficient serving with solid multilingual performance across European languages; cheap enough to translate everything, then escalate only what reads wrong.

Best for: High-volume European pairs on a budget.

Frequently asked

Run them in one workspace

Multi-Model Chat

GPT, Claude, Gemini, Grok, Llama: pick mid-conversation, cheapest-first.

Memory

A durable, private memory every agent reads and updates as it works.

Drive

Folders, previews, and lightweight versioning for AI work.

Keep comparing

The models writers actually keep.Frontier answers at commodity prices.

Judge them on your own prompt

Every pick on this page runs in one idapt chat. Start free with a daily allowance, no credit card.

Try idapt free