DeepL Translate

DeepL Translate is best judged on what it asks you to give up. Our source material records a focused catalogue of around 30+ core languages, and a design that only functions while it can reach cloud servers. In exchange it offers a wide set of ways to get text in, and tools for working on what comes out. Whether that is a good bargain comes down to two things you can check in a minute: your language pair, and how reliable your connection is where you would use it.

Camera, photo, voice, handwriting — and whole files

The input range is the most immediately useful thing our records describe: real-time camera translation, photo uploads, voice dictation and hand-drawn character input, with our FAQ adding document translation on mobile, taking compatible file uploads.

That is an app built for text you cannot easily type: a menu, a street sign, a printed letter, a page of a manual, a character someone can draw but not find on a keyboard. Handwriting input is the one that stands out — for a script you have no keyboard for, drawing it is the only practical route in.

Whole sentences rather than word-by-word

Our source material describes the design intent clearly: deep learning models trained on multilingual data, with the engine analysing the structure of an entire sentence rather than converting it word by word, to hold onto idiom, nuance and industry-specific jargon. That is the architecture as our records describe it, and it is the basis of everything else here.

We have not translated a sentence with it, so what we can add is other people’s accounts. One recurring theme in published feedback is exactly this: reviewers describing output that captures natural conversational phrasing and picks up context specific to their field. A second comes from bilingual professionals, who describe running their own messages past the app before sending them to international colleagues.

Tapping a word to change the phrasing

Alternative Translations is what makes that second theme work. Our source material records that you can tap a specific word or short phrase and see other phrasings, then pick the one that fits. It turns the app from something that hands you an answer into something you can negotiate with.

Alongside it sits DeepL Write, recorded in both our source material and our FAQ as a built-in AI writing assistant suggesting grammatical corrections, better word choices and stylistic or clarity improvements. It is not a translation tool but an editing pass aimed at text you wrote yourself.

For business and enterprise accounts, our source material records Single Sign-On login and tools to build custom glossaries, which matter when an organisation has fixed wordings for product names or internal terminology.

Around 30+ core languages

Our source material is direct about the catalogue: around 30+ core languages, weighted towards European and major Asian and global markets. Our cons list records that same shape as a limitation.

It is also the one fact here you can settle against your own needs in a minute, and it is worth settling first: if your language pair is not on the list, nothing else in this review matters. The current list lives on the live Play listing.

What a DeepL Pro subscription covers

Our cons list records character limits on the free tier for the text you translate at once, and two customisation tools behind a paid plan: custom glossaries and tone adjustment. Our FAQ narrows the second further — choosing between formal and informal tone is part of DeepL Pro, and it names German and French as examples of languages where that distinction applies.

That matters to a particular user. If you are writing to a German or French colleague and formal versus informal address is the difference between a normal message and an awkward one, formality selection is not a decorative setting; it is recorded as a paid feature. The in-app purchase screen and the Play listing are where a subscription’s current cost and coverage are set out.

The microphone stops when you pause

Dictation is the one weakness our material records twice, independently, and the two accounts match. Our cons list records that the voice tool lacks automatic punctuation and that the microphone can shut off too quickly during pauses. A recurring theme in published feedback describes the same behaviour from the user’s side: the microphone cutting out during brief pauses, leaving people speaking rapidly without stopping to avoid restarting the dictation.

Both halves matter in practice. No automatic punctuation means a dictated paragraph arrives as one undifferentiated run of words, and punctuation is not decoration in a translation — it is where the sentence boundaries are. A microphone that treats a pause for thought as the end of an utterance changes how you speak: not at your own pace, but at the pace the timeout allows. For a short phrase at a counter, fine. For anything longer, dictation here is a second-choice input rather than a first.

It does not work without a connection

This is the sharpest practical fact here, and our records state it twice. Our FAQ gives the reason: the translation models run on cloud-based neural networks, so an active internet connection is required. Our cons list states the consequence — no offline functionality, and a stable connection needed to reach the translation servers.

Note who that lands on. Our own source material names travellers among the people this app is for, and travel is where a phone’s connection is least dependable: in the air, at a border before a roaming SIM attaches, on a rural road, anywhere a data plan has run out. A desk user on office Wi-Fi will never meet this limitation. A traveller may meet it at the exact moment the app would have been most useful, which makes this a tool you plan a connection around rather than one you assume will be there.

When the app cannot see a connection

Separately from the design above, one recurring theme in published feedback describes a bug: the app showing a connection error or failing to recognise that the phone is online, blocking translation until it is reinstalled.

These are two different problems. The first is by design and predictable — no signal, no translation. The second is the app failing to see a connection that exists, with reinstalling reported as the route back.

The listing itself

Our source material places the developer, DeepL SE, in Cologne, Germany, and our records carry a Play update of Apr 3, 2026 — a little over five months before this review, on a listing whose 4.7 average rests on 391,321 ratings.

Who it’s not for

If you need translation when the network is gone, this is the wrong app, and the limitation is structural rather than a bug awaiting a fix. Dictation-heavy users will meet the punctuation gap and the microphone timeout. And if formality control or a glossary is what you came for, our records place both behind a paid plan.

Our take

We land at 4.5, two-tenths under the store’s 4.7, and the gap sits on dictation, not on the offline limit that first looks like it should carry it.

The tempting argument is the offline one. Our own source material names travellers among this app’s audiences, and the same material records that it cannot work without a connection. Those two facts do pull against each other — but that is a mismatch inside a description, not a fault inside the product. Cloud-only translation is a stated, static property of the app — our FAQ gives it as the reason the thing works at all — and it applies identically to everyone who opens it. An average built entirely under that condition has already absorbed it. Charging it now would be charging the app for the shape our own records say it has. The offline limit decides who this suits; it does not tell you how well the app translates for the people it does suit.

Dictation is a different kind of thing. Nobody choosing a translator agrees in advance that the microphone will stop when they pause for thought, or that dictated text will arrive without punctuation. That is behaviour rather than a term of purchase, and our material records it from two directions — our own cons list and a recurring theme in published feedback — which is corroboration rather than repetition. It degrades one of the recorded routes into the app that does not need a keyboard, and it is a fault no buyer could have priced in from the listing. The reported connection error sits in the same class and adds to that gap without widening it much.

The catalogue size and the Pro gating of character limits, glossaries and formality did not move the number either, and for the same reason: both are disclosed shapes of the product a buyer weighs on the store page.

What holds the score high is how much the rest of the offer covers: four ways to get text in without typing it, whole-document translation, a phrase-level alternatives tool that lets you argue with the result, and a separate writing assistant for your own prose. For anyone working between a handful of major languages on a dependable connection, that is a strong package.