Translate between 40+ languages, with no account and no page limit

Google's official translation API requires a billing account and an API key. Putting that key in a public web page would expose it to anyone who opens the developer console, and the bill would land on us.

The unofficial endpoints people scrape instead violate Google's terms of service and get blocked without warning. Neither is a foundation for a tool that should still work next month, so this page uses services that publish an open API on purpose.

5,000 characters per go, roughly two pages of prose. The counter under the input box turns amber as you approach it.

Behind the scenes the translation API accepts only 500 characters at a time, so longer text is split automatically. The tool cuts at paragraph breaks and sentence endings rather than mid-word, which is why a long passage shows a "section 3 of 7" progress message — each section is a separate request, and the results are stitched back together.

For very long documents, work in chunks of a few paragraphs. You will get a better result than pasting ten pages at once, because you can check each section as it comes back.

The provider guesses the source language from the text itself and reports what it decided — you will see it appear under the output pane.

Detection is reliable for a sentence or more, and unreliable for single words. "Pain" is a valid word in both English and French; no detector can resolve that without context. If you are translating one word, set the source language manually.

Two engines sit behind this tool. If MyMemory is rate-limited or down, the request goes to a LibreTranslate instance instead, and the two do not word things identically — they are built on different models.

The status line next to the Translate button names which engine answered, so you can tell them apart. Both are legitimate translations; they just made different word choices.

Long text adds a second source of variation. Because it is translated in sections, each section is translated without knowing what the others said — so a term may be rendered one way in paragraph one and slightly differently in paragraph four. Worth a read-through before you use it.

Not on this page — it is built for text you paste in. File translation needs to preserve formatting, which is a genuinely different problem from translating a paragraph.

For a PDF, extract the text first, translate it in sections here, then rebuild the document. It is more work, but you keep control over the layout instead of letting a converter guess at it.

First, try simplifying the source. Machine translation handles plain declarative sentences far better than long ones with nested clauses, idioms or sarcasm. Splitting one complicated sentence into two simple ones often fixes it outright.

Second, check the source language is set correctly rather than left on auto. If the detector guessed wrong, everything downstream is wrong too.

Third, accept the limit. For anything with legal or commercial consequences, machine output is a first draft for a human translator, not a substitute for one.