Document Understanding Release Notes
Release date: 11 November 2025
Document Understanding™ 2.2510 LTS Release
New roles and permissions
This release introduces a new role, Document Understanding Data Annotator. Designed specifically for document annotation, users assigned with this role cannot delete documents in order to maintain a secure data management process.
We have also enabled a new permission, Documents.Delete. Users granted this permission can delete documents. The roles with this permission include:
- Document Understanding Administrator
- Document Understanding Developer
- Document Understanding Model Trainer
For more information on roles and permissions, check the Role-based access control page.
New Project Administrator role
A new role, Project Administrator, is available at project level. Users with this role have all permissions to manage a project and users for the project.
For more information, check the Role-based access control page.
Custom roles
You can now define new roles using a combination of permissions to tailor access controls to your specific needs, offering more granular control than default roles.
For more information, check the Role-based access control page.
Continuous support for API keys
Even though API Key authentication will be removed for Automation Cloud™ services, Document Understanding™ is an exception to this. In terms of security, API key authentication in Document Understanding is as safe as Personal Access Tokens (PATs). Therefore, you can continue using API keys to access model endpoints, while keeping your UiPath resources secure.
For more information about the API key removal and the transition to PATs, visit Upcoming removals for Automation Cloud and Migrating from API keys to Personal Access Tokens (PATs).
Unified Pricing, an innovative licensing model
Unified Pricing brings new licensing plans, user licenses, and a consolidated consumption unit. We are also excited to let you know that Document Understanding is now available through Unified Pricing, and its license consumption is measured by Platform Units.
For more information on this new licensing model, visit the Unified pricing: Licensing plan framework page. For overview licensing information, refer to the Automation Cloud admin guide and the Overview guide.
Compare model
You can now compare the performance metrics of two different versions of both classification and extraction models, providing in-depth insight into their efficacy and the enhancements made over time.
This new feature simplifies the tracking of model improvements and changes over time by offering a user-friendly comparison view.
For more information, check the Compare model section from the Measure page.
Train models using data from Validation Station
You can now use documents that were validated in Validation Station to further improve the performance of your models. Visit the Retrain models how-to page for more in-depth information on how retraining candidates are collected and how to leverage these to retrain your models,
You can currently only retrain documents that were processed using the UiPath.DocumentUnderstanding.Activities package. All documents that were processed using this activity package and were validated in Validated Station are collected automatically and can be used for retraining.
Classification fields in Document type manager
Classification fields are now generally available in Document type manager. Classification fields are data points that refer to a document as a whole. For example, the expense type of a receipt (food, hotel, airline, or transportation) or the currency of an invoice (USD, EUR, JPY) are classification fields.
For more information, check the Classification fields section from the Build page in the Document Understanding Modern Projects User Guide.
Known limitation
- When using the Extract Document Data activity, classification fields are supported for modern project extractors and out-of-the-box models, but not for classic project extractors.
- Classification fields are extracted for custom document types only after a successful training.
- We have added a new Delete button in case a dataset export is failed. You can now use this button to delete the failed dataset export and try again.
- You can now change the document type for documents imported as zip files from both classic and modern projects.
- You can now reorder the annotation fields within the list. Access this feature from Document type manager.
- For an improved experience, we updated the
dmyParsingproperty when updating the type of a field in the following situations:- When changing the type of a field from
non-datetoDate, thedmyParsingproperty is set to auto. - When changing the type of a field from
Datetonon-date, thedmyParsingproperty is set to null. -
The advanced metrics file now incorporates actual names for both files and tags. This is an update where the advanced metrics file previously only included GUID values for file and tag names.
- When changing the type of a field from
- We enhanced the table annotation feature. You can now annotate and confirm individual rows within a table independently, without affecting the remainder of the table.
- Users can now exclude fields marked as missing when filtering documents by Annotated Field Name inside the Annotations Service.
- The new CPU inference feature reduces dependency on GPU resources, allowing you to run Document Understanding with as few as two GPUs. For more information, see the Additional Document Understanding modern projects requirements section.
