We’re excited to announce that a new trainable document splitter and classifier model is now available in Public Preview for tenants based in Europe and the US.
This new model extends the power of document classification to multi-document packets — enabling you to split and classify documents within a single workflow.
Key details
- Availability: Currently, the feature is only available in Europe and the US. Availability for new reagions will be confirmed at a later date
- Pricing: The new splitter and classifier model falls under the existing Modern Projects pricing model, where each page is charged 1 AI Units regardless of numbers of operations on that page. Please note that final pricing for General Availability (GA) may be subject to change.
What it does
The new model can:
- Classify entire documents (e.g., identify whether a file is an ID, invoice, or application form).
- Split and classify multi-document packets — such as mortgage or loan application files containing multiple sub-documents.
For example:
A 10-page mortgage application PDF might contain:
- Pages 1–2 → ID
- Pages 3–6 → Application form
- Pages 7–10 → Bank statement
The new model can be trained to automatically detect these boundaries and assign the correct document type to each section.
Getting Started
Create a project
- Start a Modern Project and enable the toggle “Enable new splitter and classifier model.”
- To perform splitting in addition to classification, also turn on “Enable splitting.”
Upload your documents
- Go to Classify and Split Documents and upload your document packets.
- Once processed, select the uploaded files and click Split to open the annotation interface.
- If your project already has a trained model, the documents will be pre-annotated automatically — saving time and showing predictions.
Define your classification taxonomy
- Click New document type to define each document type in your taxonomy.
- Choose from predefined types or create custom ones with:
- A name
- A short description (1–3 sentences)
- A few key indicators (e.g., invoice number, total amount, seller information)
Annotate and confirm splits
- Mark where each document starts and ends, and assign a document type to each range.
- Click Confirm to process and generate sub-documents.
- Each sub-document appears under its document type in the Build section and gets pre-annotated with the schema of that type.
- You can skip non-relevant pages by labeling them as “–Unknown–.”
Train your model
- Training begins automatically once you have at least five annotated sub-documents.
- Training status is visible in the Classification pane.
Review metrics
- Navigate to the Measure page and review model metrics
Publish the Model 
- Once training is complete, publish your model in the Publish section to make it available for use in your automations.
- Published models can be versioned, managed, and reused across projects.
- The version of the new splitter and classifier model is 25.9
Consume the Model in Your Workflow or via APIs 
- Use the published model directly in UiPath workflows to automatically classify or split incoming documents.
- Currently, the new splitter and classifier model can be consumed through IntelligentOCR.Activities 6.27.0.
- You can also access the model through APIs to integrate document processing into other systems.
Reviewing Predictions
After training, all project documents are updated with predictions from the model.
You can review results by:
- Comparing Ground Truth (Type) vs Predicted Type in the Classification table.
- Viewing sub-documents by enabling “Include sub-documents” in the View menu.
- Enabling “Show Prediction” in the annotation interface to see how the model performed.
Classification-Only Option
If you only need classification (not splitting), simply disable the “Enable splitting” toggle.
The model will then classify whole documents as before.
Current Limitations (Public Preview)
Some limitations apply during the preview phase:
Dataset
- Minimum document types: 1
- Minimum samples:
- Single document type → at least 5 samples
- Multiple types → at least 5 total documents (1 per type minimum)
- Maximum document size: 160 MB or 500 pages
- Training triggered after 5 annotation changes
Annotation
- Pages cannot be reordered or deleted
Features not yet available
- Splitting info in the Monitor page
- Retraining for splitting/classification
- Splitting support in for the cross-platform activities (DU.Activities)
- Migrating splitting and classification data sets across environments
These limitations will evolve as we move toward General Availability (GA).
Share Your Feedback!
We’d love to hear your thoughts as you try the new splitter and classifier model:
- How well does it handle your document packets?
- What improvements would help before GA?
- Any issues or surprises you encountered?
Your feedback will help us make the feature even better before full release.