What you're sending decides the AI translation API you want: short text strings, whole documents, or files that must keep their layout. Big cloud APIs suit high-volume text at a per-character price, specialist engines suit documents and quality-sensitive copy, and an open-source engine suits teams that need to run translation on their own servers.
A checkout button that reads "Abandon basket" in German costs more than any API bill. So does a refund policy where "not" went missing in Spanish, or a product name the engine helpfully translated into a common noun. Machine translation is cheap per character and expensive per mistake, which is why the choice of API matters less for its price than for how it handles your kind of text.
The forks that decide it
Text strings, documents or files. Translating a sentence from your app is a different job from translating a Word file or a PDF and getting back something with the same headings, tables and layout. Some APIs here are built for real-time text; others translate documents natively or in batches. If your content lives in DOCX or PDF and the layout must survive, shortlist services that translate those files natively; extracting the text and rebuilding the file yourself is a job nobody enjoys twice.
Hosted or self-run. A hosted cloud API is the fastest start: an account, a key and a few lines of code. An open-source engine you run on your own servers means you control where the text goes and what happens to it, at the price of running and updating the engine yourself. If your text can't leave your own infrastructure, that's the honest answer, upkeep and all. If it can, start with a big cloud API for app strings and support content at volume.
Where the data goes. If you translate contracts, medical notes or customer messages, ask each vendor where translation happens, whether text is stored and for how long, and whether it's used to train anything. Check the answer in the vendor's own documentation rather than the marketing page, and if residency matters for your sector, make it a hard filter before you compare quality.
Terminology control. Brand names, product names and legal terms need to come out the same every time. Before you commit, find out how each API lets you lock terms, whether through a glossary, a custom-trained engine or a do-not-translate list. If you have fixed terminology and plenty of your own translated material, a service with custom-trained engines will repay the setup better than a slightly stronger general engine.
Machine only, or machine plus people. At least one platform in this group connects machine output to human translators inside a workflow. If anything you publish needs a human sign-off, pick that routing from the start rather than bolting it on later.
Cloud APIs for high-volume text
DeepL API
A developer API from DeepL covering text plus Word and other documents, with a free key to start building against.
Amazon Translate
Teams already on AWS can add neural machine translation here, for real-time text or for documents sent through in batches.
Azure Translator
Microsoft's AI-powered text translation service, which sits inside its Foundry Tools for apps built on Azure.
Alibaba Cloud Machine Translation
Neural machine translation from Alibaba Cloud, driven through an API or console and paid for in character-based resource packages.
Specialist engines, open source and workflow platforms
Lara Translate
From Translated, an engine that handles text, documents, images and audio across a very wide language list.
LibreTranslate
When translation has to stay on your own servers, this open-source API can be self-hosted, or used through paid hosted keys.
Tilde Machine Translation
Text, PDFs and documents go through this service, which offers an API integration and engines custom-trained for your material.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Smartcat Translation API
Built for enterprise workflows, this API links machine and human translation, with document translators on the same platform.
Translate.com API
A translation service that lists a machine translation API next to its document translation offering for files.
A blind review before you commit
Quality scores published by vendors are measured on someone else's text. Yours is the only test that counts, and you can run it in an afternoon.
Pull fifty strings from your own product: a dozen short buttons and labels, a dozen full sentences from help pages, a handful of strings with placeholders like {name} or HTML tags, some sentences containing your brand and product names, and one or two that are deliberately awkward, such as an idiom or a negative that changes the meaning. Add one real document if documents matter to you.
Send the same set through two or three APIs into your main target language. Strip the vendor names, shuffle the outputs, and hand them to a native speaker who knows your product. Ask them to mark each string as fine, fixable or wrong, and to circle anything that would embarrass you in front of a customer. Check separately whether placeholders and tags came back intact, because a broken {name} is a bug, not a style issue.
Then open the translated document and compare it with the original page by page. The winner is the API with the fewest "wrong" marks, not the one with the most "fine" ones. If you're translating spoken content too, this round-up of AI video translation tools covers that side.
Do you make AI translation APIs or machine translation engines? Tool makers can ask to be considered for a featured place on this list.
What it costs
The cloud APIs that publish a rate charge per million characters, usually with a free allowance on top. Free tiers vary in shape: a monthly quota that renews, a credit applied to the bill, or an allowance that only lasts for your first year. Read which one you're getting before you plan a launch around it.
For standard text, the per-million-character prices re-checked on the vendors' own pricing pages this month fall between about $15 and $25. Document translation can cost more than plain text on the same service, and formats such as DOCX may be priced separately from text and HTML. Several smaller services here sell a monthly plan instead, at least one with an included character allowance and an overage rate. That's easier to budget, and poorer value if your volume swings.
A cost worth knowing about this year: at least one big cloud provider now prices its large-language-model translation at parity with its standard neural engine, so the choice between the two is about quality on your text rather than the bill. And one specialist engine has announced it will close by the end of 2026, moving users to a successor service. Check that the API you pick is the product the vendor is investing in.
A self-run open-source engine swaps the per-character bill for server costs and your own time. It can work out cheaper at steady high volume, and dearer when nobody on the team wants to look after it. If you already self-host other AI models, such as open-source speech-to-text tools, you'll know which side of that line you're on.
Mistakes that reach customers
- Translating strings without context. A button label on its own is ambiguous; send the surrounding sentence or a note where the API allows it.
- Letting the engine translate brand and product names. Lock them before launch, not after the first complaint.
- Testing only your best language. Quality varies by language pair, so test every language you plan to ship.
- Skipping the placeholder check. A translated variable name breaks the page as surely as a code bug.
- Choosing on price per character alone. A cheaper API that needs a human to fix one string in ten isn't cheaper.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.


