Reglyph is our product, so we can't claim to be neutral. DeepL is one of the best-known names in translation, and it performed strongly in this test. Same files, same target languages, every result shown as it came out.
| Page | Languages | Better result | What decided it |
|---|---|---|---|
| Text over a photo (scan) | EN → ZH | Reglyph | Background: DeepL left smudges on the icons and flattened the big numbers |
| 1871 book page (scan) | FR → EN | DeepL, slightly | Detail: DeepL kept the indents and italics |
| Colored panels (scan) | EN → ES | Tie | Our tab label overflowed; DeepL split a bullet across columns |
| Financial table (scan) | EN → JA | Tie | DeepL left patchy cell shading; we left a smudge and lost the bold |
| Dense two-column paper (digital) | EN → FR | Reglyph | DeepL's label covered the chart; we lost the indents |
| Paper with a parameter table (digital) | EN → KO | Reglyph | DeepL's file ran to two pages and split the table |
| Score | Reglyph 6 · DeepL 4 | A tie scores a point for each |
Unlike the other tools in this series, DeepL never laid a solid backing under its translation and never skipped a table. Most of what separates the two is detail.
On scans DeepL is the strongest tool we've tested so far. It read all four image-only pages, translated everything, and kept the backgrounds. The gap shows up on the busiest background, and in the small typographic details where DeepL is more careful than we are.
This is the page with the hardest background, and the one where we pull ahead. Look at the band of three statistics. Behind DeepL's text, the briefcase and the banknote are smeared where the English used to be, and the pie chart has a square patch cut into it. The big bold numbers became ordinary lines of text, and the 40% figure lost "in coastal counties". Ours restored the icons and kept 40%, 54.6 million and $10 trillion as large display numbers. They aren't perfect either: the original numbers are bold, ours are thin.
Both kept the old paper and the engraving, and both translated the running header. DeepL wins on the details: every paragraph keeps its first-line indent, and the book title at the bottom is still in italics. Ours lost the indents, so the paragraphs run together. DeepL has two small slips of its own, "frigateAbraham" with no space and an asterisk in front of the title, but its page reads more like the book.
A tie. Our "Salud intestinal" tab label runs onto two lines and leaves a smudge in the blue tab, and the first heading wraps. DeepL's tab is clean, but its dairy section breaks one bullet in the middle of a sentence, with the end of it at the top of the next column. Our body text is a little easier to read; DeepL's is closer to the original typeface.
Another tie. Both translated the whole table, the footnotes and the chart, with every number where it belongs. DeepL kept the bold in the headers and totals, which we didn't. But one of its rows, the Ocean Energy line, has white blocks breaking up the row shading. Ours has a small smudge in the footnotes.
Digital PDFs are DeepL's home ground, and its running text is more polished than ours: justified lines and first-line indents. But both papers also had a figure or a table, and that's where it slipped.
Both tools fit the French onto the page. DeepL's columns look more finished, with justified lines and indents, where ours have no indents. But DeepL also went further than we did and translated the labels inside the figure, and that backfired: "Relativité numérique" came out several times too large and now sits across the top chart, and the waveform it labels can no longer be seen. That figure is the paper's main result. Ours is untouched, with its labels still in English, so this page goes to Reglyph.
The original is one page. DeepL returned two. The body text is well translated, but the extra length pushed the parameter table off the first page, and on the second page the table is split into two pieces, half on the left and half on the right. Here is the whole file:
DeepL also left the three figure captions (Vanilla Attention, Sliding Window Attention, Effective Context Length) in English, and translated some of the table's configuration keys, so hidden_dim became 숨겨진 차원. Ours kept everything on one page, translated the captions and left the keys as written.
The two tools charge differently. DeepL counts files: its pricing page lists 1 file translation a month on the free plan and 3 a month on Individual, the cheapest paid plan. Reglyph counts pages: 5 free pages every day, $0.30 a page with no subscription (minimum $3), or $9 a month for 100 pages.
Which is cheaper depends on what you translate. One long document a month fits DeepL's file count easily. Short ones add up faster: a contract, a certificate and two scanned letters are already four files, one more than Individual allows. That's where paying per page works out better.
More side-by-side tests: Reglyph vs Reflo on the same six pages, Reglyph vs Google Translate, Reglyph vs ChatGPT, Reglyph vs Smallpdf, and the best PDF translation services for a wider roundup.
Upload a scanned PDF — even one made with your phone's scanner app. Reglyph OCRs it, translates it, and rebuilds the page so tables, stamps, and figures stay exactly where they were.
Translate 5 pages free, every dayYes. In our test it read four image-only scanned pages, translated all the text and kept the backgrounds. It was weaker on a page with text over a photo, where it left smudges behind the statistics and flattened the large numbers.
Mostly. On five of our six pages it kept the layout close to the original, including first-line indents, italics and bold. On a Korean research paper it returned two pages for a one-page original and split the parameter table across them, and on a French physics paper an oversized translated label covered a chart.
It was close. Reglyph did better on a page with text over a photo; DeepL did slightly better on an old book page thanks to its indents and italics; the other two scans were ties.
It depends on how many documents you translate. Reglyph charges by the page: 5 free pages a day, $0.30 a page without a subscription, or $9 a month for 100 pages. DeepL counts files instead, so one long document a month may cost less there, while many short documents usually cost less with Reglyph.
We can't be neutral, so we kept it checkable: the same files and target languages for both tools, every result shown next to the original as it came out, and the pages DeepL won included. All six source pages are public, so you can run the same test yourself.
Vision models genuinely read scans now. The question isn't whether ChatGPT can translate your PDF — it's whether a chat reply is actually what you need back.
Read guideCamera-translate apps are great for reading a page on the spot, but they don't give you a file — here's how to get an actual translated PDF from your phone.
Read guideRandom symbols, blank pages, or 'PDF tools are temporarily offline' — the cause is almost always the same one thing, and it's fixable.
Read guideSimple, scan-friendly pricing. Pages are pages — no multiplier for scanned files.
Translate a few pages. No commitment.
$0.30 a page, from 10 pages — you pay for exactly 10 pages.
For individuals with scans to translate every month.
For high-volume users with scans to translate every week.