Ukrainian and Russian text from images - 5 Apify OCR Actors compared, only one read the invoice table
"Image to text" is one of the most searched tasks in the Apify Store, and most OCR Actors are tuned for English.
If your images are Ukrainian or Russian - screenshots, photos of documents, scanned invoices - the results differ a lot. I ran five OCR Actors on the same three images and compared every output with the text the images were made from. All results are from real runs on 16 November 2026.
Disclosure: one of the five is mine, Image to Text OCR. Every number below comes from a real run, the images are public and the raw outputs are kept.
Short answer
| Actor | Ukrainian text | Russian text | Invoice: numbers exact | Invoice table | Price per image |
|---|---|---|---|---|---|
ai_admin/image-to-text-ocr |
100% | 100% | 12 of 12 | rows, exact | $0.004 |
zen-studio/google-lens-ocr |
100% | 100% | 11 of 12 | read column by column | about $0.007 |
getascraper/google-lens-ocr |
100% | 100% | 10 of 12 | read column by column | about $0.004 |
steadydata/image-ocr-text |
Latin gibberish | Latin gibberish | 7 of 12 | no | about $0.004 |
gochujang/image-ocr-extractor |
error: no Ukrainian | error | - | - | $0.003 |
The three images
| Image | What it is |
|---|---|
image-uk.png |
Ukrainian heading and two sentences, clean |
image-ru.png |
the same in Russian |
invoice-noisy-uk.png |
a scanned Ukrainian invoice with noise and a slight tilt: a 5 x 6 table without lines, amounts like 40 530,00 and 2 500 |
They are public: https://api.apify.com/v2/key-value-stores/FloomuI2DrQCJazY8/records/<file name>.
Settings: my Actor with languages: ["ukr", "rus"] - one setting for both languages, as a buyer with mixed documents would use it; the Google Lens Actors with their language field set to uk or ru; the others have no language option or take a Tesseract code (ukr+rus).
The Actors are the most used ones for "image to text" and "OCR" in the Store on that day: steadydata has 178 users a month, the two Google Lens Actors 63 and 57, gochujang 32.
What each one returned
Clean Ukrainian and Russian text. My Actor and both Google Lens Actors returned the text character for character, including «і», «ї», «ё» and «ы».
My Actor does it with a single setting for both languages: it finds out which language each page is in, reads it with that language first and reports the switch in the languagesUsed field.
steadydata returned Cyrillic as Latin look-alikes - the Ukrainian heading «Звіт про постачання» came back as 3BiTnpOnOCTayaHHA.
gochujang failed on every image with Failed loading language 'ukr': the Ukrainian model is not installed, but the start of the run is still charged.
The noisy invoice
This is where the difference shows. My Actor returned the table as rows, with every number as printed, in the markdown field and as cells in tables:
Рахунок-фактура № 47
| № | Товар | Од. | Кількість | Ціна | Сума |
|---|---|---|---|---|---|
| 1 | Цемент М500 | мішок | 420 | 96,50 | 40 530,00 |
| 2 | Арматура 12 мм | шт. | 180 | 412,00 | 74 160,00 |
| 3 | Пісок річковий | т | 12 | 650,00 | 7 800,00 |
| 4 | Цегла рядова | шт. | 2 500 | 9,80 | 24 500,00 |
Разом: 146 990,00 грн
The Google Lens Actors read every word, but the table came back column by column, so the rows of the invoice are no longer together. This is the start of the getascraper output: № Товар Цемент М500. Арматура 12 мм Пісок річковий T Цегла рядова Од. мішок 2 шт.
On top of that, 40 530,00 became 40.530,00 in both, the unit «т» became a Latin T, and getascraper split 24 500,00 into two lines. To get a table back from that, you would need another tool.
steadydata got 7 of the 12 numbers and none of the Cyrillic words right.
Price
| Actor | One image | Notes |
|---|---|---|
| mine | $0.004 | plus $0.0001 per run |
getascraper |
$0.004 | |
steadydata |
$0.004 | $0.003 read + $0.001 check, plus a start fee per run |
zen-studio |
about $0.007 | $0.005 per image plus $0.002 per run, and one run takes one image |
gochujang |
$0.003 | plus $0.001 per run |
So the price is the same as the Google Lens and steadydata Actors - the difference is what you get for it.
Verdict
- Exact text in Ukrainian and Russian, with one setting - mine and the two Google Lens Actors.
- Invoice as a real table with exact numbers - only mine.
- Price - the same $0.004 as
getascraperand steadydata, less thanzen-studio.
While preparing this article, the test results showed me where my Actor could become the best in its category - mixed Ukrainian and Russian documents and invoice tables without lines - so I improved it right away, before publishing. The latest runs described above confirm it.
Try it
from apify_client import ApifyClient
client = ApifyClient("<YOUR_APIFY_TOKEN>")
run = client.actor("ai_admin/image-to-text-ocr").call(run_input={
"sources": [{"url": "https://api.apify.com/v2/key-value-stores/FloomuI2DrQCJazY8/records/invoice-noisy-uk.png"}],
"languages": ["ukr", "rus"],
})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
print(item["markdown"])
print(item["tables"])
It is also a tool for Claude, Cursor and other AI agents - see MCP server for document and media tools.
For whole PDFs, Word and Excel files, see 5 document-to-Markdown Actors on 7 files.
Which documents do you need to read from images - receipts, invoices, screenshots? Tell me in the comments, and I will test those next.
Originally published at andriyvk.github.io.
Comments
No comments yet. Start the discussion.