Granite Guardian 4.1 8B
Granite Guardian 4.1 8B is a language model that judges prompts and responses. Built from Granite 4.1 8B, it handles predefined risk categories and custom criteria written in natural language.
Model design and capabilities
Its checks include jailbreaks, profanity, grounded answers and tool-call hallucinations. Custom criteria can cover formatting, length and domain requirements. It produces judgments for another application's content; it is not the general assistant that generates that content.Granite Guardian 4.1 8B model card
Practical use
Practical guidance. Calibrate each criterion on real acceptable and unacceptable examples. Track false positives and missed violations separately. Use ordinary code for exact rules such as required fields where possible, and reserve the model judge for decisions that need language interpretation.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | ibm-granite/granite-guardian-4.1-8b |
| License | Apache 2.0Granite Guardian 4.1 8B model card |
| Parameters | 8BGranite Guardian 4.1 8B model card |
| Custom evaluation | Bring Your Own Criteria, including multiple constraints.Granite Guardian 4.1 8B model card |
Related models
- Granite 4.1 8B
IBM · Granite 4.1
An instruction-tuned 8B text model for retrieval-based answers, extraction, coding and tool use.
Launch date API · Downloadable · Provider product - Granite Switch 4.1.1 8B
IBM · Granite Switch
A modular Granite 4.1 8B system with selectable decision and Guardian adapters, currently at revision 4.1.1.
Launch date not confirmedDownloadable