deGerman
Available94,4% strict-span F1
Documentation
Languages and data categories
See what ShinrAI 1.3 recognises, how coverage is measured and which API surface to use.
15 model locales
One multilingual model, with regional Portuguese variants and Hebrew in beta. Website languages and API language codes are separate.
No matching languages or categories.
enEnglish
Available97,7% strict-span F1
jaJapanese
Available98,1% strict-span F1
frFrench
Available96,0% strict-span F1
esSpanish
Available96,9% strict-span F1
itItalian
Available95,0% strict-span F1
plPolish
Available90,9% strict-span F1
pt-PTPortuguese · Portugal
Available94,2% strict-span F1
pt-BRPortuguese · Brazil
Available97,3% strict-span F1
ruRussian
Available96,4% strict-span F1
ukUkrainian
Available86,7% strict-span F1
trTurkish
Available90,1% strict-span F1
arArabic
Available87,9% strict-span F1
koKorean
Available95,5% strict-span F1
heHebrew
Beta59,5% strict-span F1
Business and clinical letters; 200 texts per locale. Strict entity spans. Hebrew beta is included. Model-only evaluation by Innovius, not a service-wide accuracy guarantee.
View methodology and results →Model locales, API codes and website languages
The website is available in English, German, Japanese, French and Spanish. This does not limit the model to those five languages.
The documented Azure text adapter accepts these language codes:
de en fr es it pl pt ru uk tr ar he ja ko
Portuguese uses one API language code but has two evaluated regional model locales. An accepted code does not guarantee every entity category or equal quality.
Read the adapter limits →19 data categories
Examples are illustrative. Detection depends on the surrounding text and configuration.
People and organisations
People
Emma WeberPERSONOrganisations
Example GmbHORGContact and location
Cities
BerlinCITYStreet addresses
Musterstraße 12STREETEmail addresses
emma@example.comEMAILPhone numbers
+49 30 000000PHONEPostal codes
10115POSTAL_CODEFinance
Bank accounts
DE89 3704 0044 0532 0130 00ACCOUNTCrypto wallets
0x…WALLETPayment cards
4242 4242 4242 4242CARDPersonal identifiers
Dates of birth
14.03.1985DOBNational identifiers
Tax / identity / social-security IDNATIONAL_IDVehicle plates
B AB 1234PLATECustomer and case IDs
CUSTOMER-0042CUSTOMER_IDDigital and confidential information
Usernames
example\emmaUSERNAMEWeb addresses
https://example.comURLNetwork addresses
192.0.2.1NETADDRSecrets and access tokens
API_KEY_EXAMPLESECRETProject codenames
Project AuroraCODENAMEWhy 19 categories and 27 classes?
The model has 19 detection heads. People, cities, streets and organisations each have three subtypes; the other 15 categories have one. That gives 27 classes, not 27 unrelated data categories.
PERSONcommon, uncommon, rareCITYmajor, medium, smallSTREETgeneric, specific, localORGinternational, national, regional
Origin, name form and other attributes help choose appropriate replacements. They are additional attributes, not additional languages or guaranteed detections.
Enterprise detectors and API-specific coverage
Enterprise adds deterministic rules and validation for supported identifiers. These are not included in the model-only scores above.
Adapter examples include US social-security numbers, bank routing numbers, SWIFT codes and vehicle identification numbers. Coverage and required context vary by API.
Compatibility adapters expose their documented categories and operations. Unsupported options are rejected; unmapped findings can produce warnings. Use the native API when you need findings outside an adapter mapping.
Native API and compatibility documentation →