extractingbench

Formats and languages

Does it read 03/04 and 1.234,56 the right way, and non-English documents?

Locales: US, UK, German, French and Japanese conventions for dates, numbers and currencies, on the same records and the same answers.

Languages: German, French, Spanish, Japanese and Chinese documents, plus Indonesian receipts.

Accuracy across locales worst locale

Clairic
Cloudflare
NuExtract3
NuMind
GLiNER2
Fastino

Accuracy on the same records rendered with US, UK, German, French and Japanese date, number and currency formats. The chart shows the worst locale.

Non-English accuracy German, French, Spanish, Japanese, Chinese

Clairic
Cloudflare
NuExtract3
NuMind
GLiNER2
Fastino

Accuracy on documents in German, French, Spanish, Japanese and Chinese, pooled.

Test items and precision

Items
100 records in five locales; 40 documents in each of five languages
Precision
About ±4 points per locale; about ±8 per language, ±4 pooled.

Full method and scoring rules