Amazon has updated the text-detection model in Textract, adding support for more than 40 languages and improvements to handwriting recognition and multilingual tables. The upgrade is automatic, so existing applications can use it without changing their API requests or response schema.
AWS AI Watch analysis
What happened
The updated model applies to Textract’s DetectDocumentText and AnalyzeDocument APIs and is available in every AWS Region where Textract is offered, according to AWS. The company says it improves recognition of long words and handwriting, and handles merged, skewed, rotated, partial or corrupted tables more robustly.
AWS says the expanded language coverage includes scripts that were not previously supported. The service processes scanned documents such as invoices, forms and identity documents, extracting text and structured data.
Why it matters
For teams processing documents across languages, the update could reduce the need to route unfamiliar scripts or awkward tables through a separate workflow. That is useful operationally: the model changes, while the application interface stays put.
AWS describes the gains, but the announcement does not provide comparative accuracy figures. The practical test will be how the improvements hold up on the documents each team actually receives, which are rarely as tidy as the examples in a product announcement.
Our read
This is a meaningful, low-friction improvement to a machine-learning service people may already rely on. If you use Textract, there is no API migration to plan; it is worth checking whether the new language and table handling improves your results on real documents.
What to watch
- Whether AWS publishes comparative accuracy results for the updated model.
- How the expanded language support performs on real-world documents and handwriting.
- Whether customers see fewer extraction errors on complex or damaged tables.
Discussion spark: Would you trust an automatically updated document-extraction model on important paperwork without first testing it against your own documents?
Sources and evidence
- Announcing an updated text detection model for Amazon Textract (2 November 2026, 08:00 UTC)
not affiliated with or endorsed by Amazon Web Services (AWS)